RNA-Seq Data Analyses and Machine Learning Predict Frailty Classifier Genes

RNA-Seq Data Analyses and Machine Learning Predict Frailty Classifier Genes | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article RNA-Seq Data Analyses and Machine Learning Predict Frailty Classifier Genes Marshall Tabetah This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7475052/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract A set of frailty classifier genes is constructed based on a RNA-seq dataset with phenotypically contrasting frail and non-frail samples. Authentication of the obtained classifier genes as a generic set that will be applicable to new test samples is verified by testing whether frail and non-frail samples are separately identified in a test dataset. Verification of the gene selection approach is performed by using both a supervised learning approach, k-nearest neighbor (kNN), and an unsupervised learning approach, k-means, to classify test samples, as the different approaches show remarkable coincidence in some results. The numbers of classifier genes that closely reproduce the classification outcomes based on entire gene sets are remarkably small, only ~ 10 genes. The classifier genes are shown to correlate with several notable genes in the literature. An online search of gene sets that overlap with the top 500 relevant genes to frailty reveals gene ontologies that associate with frailty. Health sciences/Biomarkers Biological sciences/Computational biology and bioinformatics Biological sciences/Genetics frailty phenotype genotype classifier genes relevant genes correlation training samples validation samples machine learning k-means test samples misfits clustering classification Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 INTRODUCTION RNA sequencing is a cost-effective high throughput sequencing (HTS) technology that sequences the whole genome with remarkable speed, throughput and accuracy [ , , ]. RNA-seq is used to quantify gene expression levels, and this has enabled its importance and application in the diagnosis and early detection of disease or pathological conditions. RNA-sequencing samples are characteristic of a high number of genes, typically in thousands or tens of thousands. However, only a subset of the genes may be related to a specific condition, making the remaining genes irrelevant. Among the relevant genes are genes that exercise some correlation across the samples in relation to the specific disease or condition. Genes that weakly correlate with one another can be used to accurately classify the condition. The remainder of genes that do not survive among the weakly correlated are thus deemed redundant. The selection method thus involves filtering out genes that are irrelevant, and then removing any redundant genes, in order to end up with a small set of genes that can adequately classify the pathological condition of interest. Obtaining a small set of genes from the sequenced gene-set is called dimension reduction in the literature [ , ]. In this perspective, genes are inherently seen as independent features, implying that the expression of a gene is in a dimension that is unique from the remaining genes. Reducing the genes to a small subset is thus resultant in a dramatic reduction in the number of dimensions initially available. Models that identify biomarker genes for predicting the prevalence of disease have been developed, as a few examples can be named: a RNA-seq transcriptomic risk score for predicting asthma, based on 73 differentially expressed genes (DEG) using Logistic least absolute shrinkage and selection operator regression analysis (Lasso) [ ]; the use of canonical machine learning (ML) classifiers such as k-Nearest Neighbor (kNN) and random forest (RF), and 1-D convolutional neural network (CNN), to predict colon cancer [ ]; the use of a gene feature selection approach for dimensional reduction to a small number of classifier genes, and subsequent application of ML classifiers kNN and support vector machine (SVM) to classify sample data [ ]. Feature selection is either an integral part of, or coupled with, the predictive model in some approaches, as in Refs [6, 7]. In other approaches, the feature selection is independent of the classification procedure that determines the phenotype, as in Ref. [8]. Another popular category of ML classifiers are clustering methods, which are unsupervised learning techniques unlike the aforementioned techniques which are categorized as supervised learning techniques. A clustering method that is implemented for classification in this work is k-means – one of the most used clustering algorithms [ ]. In this work, a feature reduction algorithm called ReliefF [5, ] is used to determine genes that are relevant to frailty in a dataset of frail and non-frail samples. Redundant genes among the relevant genes are removed using distance correlation. The genes that have survived irrelevance elimination and redundancy elimination are termed classifier genes. The subset of samples that are used to obtain the classifier genes are training samples , while the remaining samples of the dataset are validation samples . kNN is used as a classifier method to test the validation samples based on the classifier genes. An alternative method (unsupervised learning approach) used for testing of samples is k-means clustering. k-means is used for verifying whether samples of contrasting phenotypes cluster in separate groups. k-means is indispensable for testing samples in test datasets that are not the same as the training and validation dataset. kNN or an alternative supervised learning approach cannot be used for testing across datasets without account of batch effects. The data mining pipeline that involves the different stages of data processing, data analyses, and machine learning for gene selection and sample classification is displayed in Fig. 1 . There are a number of interesting findings from the investigations. The outcome of classification of validation samples based on classifier genes in a supervised learning approach and in an unsupervised learning approach can show up to be the same. With only ~ 10 classifier genes, the classification of samples closely reproduces the outcome of classification with the entire gene-set made of tens of thousands of genes: classification with a size reduction factor of more than 1000 is thus achieved. Classification across a test dataset of frail and non-frail samples by k-means shows that there is no clear division between frail and non-frail samples. The picture provided by the k-means clustering is similar to the distribution of samples in a multi-dimensional scaling plot that shows frail and non-frail samples interspersed with one another. When a 4-centroid k-means approach is adopted, most of the samples in any cluster are of the same phenotype, making the belonging of the test samples more discernible. RESULTS AND DISCUSSIONS A dataset of human frail and non-frail samples leveraged from the Gene Expression Omnibus (GEO) database is used for determining classifier genes of frailty, by using a sub-set of the data to train the gene selection algorithm, and the remainder of samples to validate the classifier genes. The dataset is a table of raw counts data made of 80 samples that is read into RStudio 4.3. Among the 80 samples there are male and female samples. Only groups of young males, frail males, young females, and frail females are of interest here, as the author wishes to properly distinguish between non-frail (represented by young samples) and frail samples. A subset of all young and frail samples is used as training samples for performing the selection of genes that are uniquely relevant to frailty. The select genes are later used to test whether a newly introduced sample from the remaining subset of samples is frail or not. The remaining subset of samples that are used to test frailty are called validation samples. Testing for a condition within the same dataset of samples is called validation. Data Processing: Identification of Outlier Samples, Outlier Gene Counts, and Normalization of Gene Counts The training dataset is composed as a set of two phenotypically distinct groups of young and frail samples. There is a total of 24 frail samples and a total of 26 young samples, making a total of 50 samples. After removing any outlying samples, the surviving samples are split into a training set and a validation set. The script DESeq2.R, as described in Ref. [ ], is used as a reference to run code that detects any outlier counts and performs normalization of the gene counts in the dataset of 50 samples. After genes with a sum of counts less than 10 across all samples are removed, the number of surviving genes is 24993. Fifty-one (51) gene counts are identified as outliers, and replaced by modified counts. Identification of Outlier Samples based on visualization of MDS plots The essence of detection and removal of outlying samples is to remain with two groups of samples that are not only phenotypically distinct, but also close to being genotypically distinct, to enable the selection of genes with expressions that show a proper contrast between the two groups. As a prior attempt to visibly identify outlying samples, multidimensional scaling (MDS) is applied to the dataset of normalized counts to see the similarity of samples by their leading log Fold Change (logFC) in a MDS plot. The distances between samples correlate with the dissimilarity between samples in a MDS plot. Samples that seem to be visibly closer to samples of the phenotypically contrasting condition than to samples of the same phenotypic condition are potential outliers. By visualization of MDS plots, the frail samples deemed to be outliers and removed are s326_M_Fr , and s323_M_Fr ; and for the young samples – s525_F_Yo , s514_F_Yo , s508_M_Yo , and s530_F_Yo . The MDS plot of samples after removal of extreme outliers s525_F_Yo and s326_M_Fr is displayed in Fig. 2 . Most of the frail samples are located in the upper left, while most of the young samples are located in the lower right part of the plot. See methods section that demonstrates samples s525_F_Yo and s326_M_Fr as extreme outliers. Identification of Outlier Samples based on k-means After implementing k-means clustering, samples that do not cluster with their phenotype are potential outliers. After removal of outlying samples, the clustering algorithm is applied again to the remaining samples to check whether all the samples cluster only within their phenotype, and the clustering is repeated until all the samples only cluster within their phenotype. Reducing the samples to two groups of genotypically distinct samples that represent phenotypically distinct groups is of interest to the author on the outcome of disease prediction. A small experiment is run to show the importance of using a MDS plot to remove samples that evidently seem distant from the rest of samples of their phenotype, or samples that clearly mingle within the contrast phenotype. It is shown that the described samples can significantly shift the centroid of one cluster, leading to several samples not being categorized in their parent group. The samples that are removed after the repetitive clustering that leaves only samples that cluster with their phenotype are: s525_F_Yo , s514_F_Yo , s508_M_Yo and s530_F_Yo among young samples; and s326_M_Fr , s323_M_Fr , s315_M_Fr , s317_M_Fr , s318_F_Fr , s302_M_Fr , s303_M_Fr , and s306_M_Fr among frail samples. With the remaining surviving samples, in all 12 runs of the k-means algorithm, all the phenotypically young samples cluster in one category, and all the phenotypically frail samples cluster in one category! The dataset of 22 young samples and 18 frail samples are deemed robust in their phenotypic and genotypic classifications, and are thus used to train and validate the gene selection. The selected samples are split into a training group made of 11 young and 11 frail samples, and a validation group made up of 11 young and 7 validation samples. As a remarkable expectation, all the samples that were visible outliers from MDS plots are among the samples that did not survive the clustering into two phenotypically distinct groups. Selection of Gene Features The ReliefF algorithm is a feature selection method that can effectively eliminate irrelevant features, by removing features with weights that are below a given threshold \(\:\delta\:W\) , leaving a feature subset that is relevant to distinguishing different types (or groups) of samples [5]. Features here are genes. The weight of a feature is determined by its ability to distinguish short-distance samples of contrasting phenotypes. A randomly chosen number of samples \(\:m\) is used to perform selection, and for each sample, there is a given number ( \(\:k\) ) of nearest samples hits and misses to be considered. Although \(\:k=10\) is used in many investigations as the most common assumption [10], a value for \(\:k\) lesser than that is used. With up to 11 training samples per phenotype, the maximum value for \(\:k\) can only be 10. Adopting the maximum value does not allow for a significant variation in the nearest neighbor samples for each target sample. The author is interested in the outcome of this variation on classification. The implemented scheme here is as in Ref. [5]. The algorithm of implementation is described in the methods section, and is coded in RStudio. The selected relevant genes are ranked starting from the most relevant to the least relevant. The feature selection does not take care of redundancy among the features, and so a correlation algorithm has to be applied in order to eliminate redundancy. The relevant genes are provided as input to the distance correlation algorithm, in order to remove genes that are redundant, so that the selected genes weakly depend on one another. The counts of a gene across samples constitute a vector of numbers. The aim in distance correlation is to determine the level of dependence of two vectors, denoted as \(\:R\) . Distance correlation satisfies \(\:0\:\le\:R\:\le\:1,\) and \(\:R=0\) only if vectors \(\:X\) and \(\:Y\) are independent [ ]. The distance correlation is evaluated by the function dcor( \(\:X,Y\) ) which is provided by the energy package in RStudio. Genes are determined to be correlated beyond a threshold dependence limit \(\:{R}_{th}\) . Distance correlation is preferred here over other correlation schemes such as Pearson correlation due to its ability to capture both linear and non-linear relationships between vectors, as discussed in Refs. [ , ]. The cut-off values, \(\:\delta\:W\) and \(\:{R}_{th}\) , that define the selected set of classifier genes are determined based on the number of classifier genes desired. The aim is to end up with a small set of classifier genes, ~ 10, that still closely reproduces the classification tendencies in validation and test samples. The correlation between two genes is tested over frail samples only. The author projects that including both samples of contrasting phenotypes will provide a sharp change in the correlation behavior at the transition between samples of contrasting phenotypes. Such behavior will disrupt the true correlation of genes among samples of the phenotype that characterizes the condition of interest – in this case, frailty. A number of relevant genes survive as uncorrelated after running several iterations in which a surviving gene is tested for correlation against all other genes of lower weight in relevance to frailty. The select uncorrelated genes can be called classifier genes, as it is adequate to classify for frailty based on them. It is demonstrated in the following sub-sections that classifier genes that are obtained from the surviving MDS plot samples are adequate in classifying test samples, with a score of 5/7 for young validation samples, and a score of 6/7 for frail validation samples, based on kNN classification. A slightly better accuracy is obtained with classifier genes obtained based on surviving k-means samples, with a score of 10/11 for young validation samples, and a score of 5/7 for frail validation samples. Computation with samples (22 in each phenotypic category) selected from visualization of MDS plots at \(\:\varvec{\delta\:}\varvec{W}\) = 0.01 In the selected dataset that is made of 22 samples in each category of frail or young samples after removal of outlier samples evident from the MDS plot, 58238 genes survive after duplicate ensemblID genes are removed from the original 58283 genes. After stripping off genes that have a total count less than 10 across all samples, 24591 genes are remaining. Any gene outlier counts (67 counts) are replaced prior to normalization. Annotation of the genes is performed to find 24253 genes in the biomaRt database. And after taking off duplicate gene names, the number of surviving genes is 20638. The samples are split into 30 training samples (15 in each phenotype) and 14 validation samples. The parameters that define the selection of relevant features are \(\:m=20\) , \(\:k=9\) , and \(\:\delta\:W\) = 0.01. 10474 genes prevail as relevant to frailty. In distance correlation, 37 genes prevail as weakly correlated features at \(\:{R}_{th}=0.4\) . Performing k-means for the training and validation samples, based on the 37 classifier genes, consistently, 1/22 frail sample ( s306_M_Fr ) only clusters among all young samples, while 3/22 young samples ( s512_M_Yo , s527_M_Yo , and s528_M_Yo ) cluster among all frail samples. The representation provided by the classifier genes seems quite satisfactory. The number of iterations required to attain convergence is typically less than 5 in all 12 independent runs. The graph indicating the cluster of the samples based on the 37 classifier genes only is depicted in Fig. 3 − 2 , and a similar graph for the entire set of 20638 genes with normalized counts is displayed in Fig. 3 − 1 . The sample misfits (not clustering with the rest of their phenotype) for the clustering based on the entire gene-set are s302_M_Fr , s303_M_Fr , s306_M_Fr , and s318_F_Fr . When looking at sample misfits – samples that do not cluster with the bulk of their phenotype, there are 4 misfits out of 44 samples in the clustering that is based on the entire gene-set, and also 4 misfits in the clustering that is based on the classifier genes only. There is one common misfit in either dataset – s306_M_Fr , resulting in 3 uncommon misfits. Based on k-means, the classification score of the classifier genes relative to the entire gene-set can be obtained as in Eq. (1) , where \(\:{n}_{mis}\) is the number of misfits that are dissimilar to both datasets, and \(\:{N}_{samples}\) is the number of dataset samples. $$\:\left(1-\:{n}_{mis}/{N}_{samples}\right)\times\:100\%\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(1\right)$$ The percentage classification score of the classifier genes relative to that of the entire gene-set, after applying \(\:{n}_{mis}\) = 3, and \(\:{N}_{samples}\) = 40, in Eq. (1) is thus 92.5% . A validation dataset that is composed of the uncorrelated (classifier) genes only in the validation samples is produced. Each validation sample is tested for frailty by calculating the Euclidean distance between the validation sample and an equal number of training frail and young samples that are randomly chosen. The validation sample is classified into the phenotype of the sample that is closest in Euclidean distance to it. In the present scenario, all the training samples (30) are used to test the validation samples. Classification results show that the same samples that did not belong to their parent groups in k-means clustering of training and validation samples do not also classify by their phenotype. This is a remarkable coincidence in results of the supervised learning approach, kNN classification, and the unsupervised learning approach, k-means. The validation samples are: s527_M_Yo , s528_M_Yo , and s306_M_Fr . This means 5/7 young validation samples classify by their phenotype, and 6/7 frail validation samples classify by their phenotype. For each validation sample, a table of training samples and their distance from the validation sample are produced. Computation with samples only that survive in their phenotypic group in k-means clustering (22 young samples and 18 frail samples) at \(\:\varvec{\delta\:}\varvec{W}=0.01\) In the selected dataset of 22 young samples and 18 frail samples that survive after maintaining only samples that consistently cluster in their phenotypic group in k-means clustering, 58238 genes survive after genes that have duplicate “ensembl ID” names are removed from the original 58283 genes. After stripping off genes that have a total count less than 10 across all samples, 24238 genes are remaining. The normalized dataset is submitted to k-means clustering. In all 12 independent k-means runs, all the frail and young samples are grouped separately, and in typically 3–4 iterations. The clustering of the frail and young samples into separate categories is illustrated in Fig. 4 − 1 . Annotation of the genes is performed to find 23905 genes in the biomaRt database. And after taking off genes with duplicate gene names, the number of surviving genes is 20411. The 40 samples are split into 22 training samples (11 in each phenotype) and 18 validation samples (11 young and 7 frail). Nearest misses and hits for each training sample are found. 18 random samples out of 22, and 7 nearest sample hits and misses, are used to perform selection of relevant genes to frailty based on the criterion \(\:\delta\:W=0.01\) . 9620 genes have prevailed as relevant to frailty. In distance correlation across samples of the frail category, 54 genes prevail as weakly correlated features at \(\:R\:\le\:0.4\) . Using the dataset that is inclusive of training and validated samples with classifier genes only, k-means converges to classify all 18 frail samples in one cluster, and all 22 young samples except s516_F_Yo and s522_F_Yo in one cluster, consistently in 12 independent runs. The described outcome of the k-means clustering of the frail and young samples is illustrated in Fig. 4 − 2 . Sample s516_F_Yo is a validation sample and sample s522_F_Yo is a training sample. The classification score of classifier genes relative to the classification of the entire gene-set, based on Eq. (1), is 95% . A validation dataset that is composed of the classifier genes only for the validation sample is produced. Each validation sample is tested for frailty by calculating the Euclidean distance between the validation sample and an equal number of training frail and young samples that are randomly chosen. In the present scenario, all the training samples (22) are used to test the validation samples. For each validation sample, a table of training samples and their distance from the validation sample is produced. Classifying a sample by the phenotype of the closest sample in the table: all the 11 young validation samples classify as young except s516_F_Yo ; among the 7 frail validation samples, only samples s322_F_Fr and s328_F_Fr have a young sample as nearest sample. However, when the average distance of a validation sample from frail samples is compared to the average distance from young samples, all frail samples are validated as frail, and all young samples are validated as young, equivalent to k-means classification based on the entire gene-set! Classifying validation samples by the phenotype of the closest-by-distance training sample in the dataset made of samples that have survived outlier detection using k-means, 10/11 young samples and 5/7 frail samples are properly validated. The dataset that is composed of samples that survived outlier detection by using k-means is used for further analyses that involve selection of classifier genes and testing of samples for classification in this work. This dataset boasts of slightly better results than the dataset with samples that have survived MDS plot visualization, although classification scores with either dataset are very satisfactory – above 90%. The excellent scores have been obtained with classifier genes that are above 30 in number. An outcome with only about 10 classifier genes is of interest to the author. This requires tightening the feature selection criteria. In processing attempts to obtain ~ 10 classifier genes, it is noticed that most of the surviving genes are those with very low counts in the training samples – less than 1 count per sample on average. Typically for these genes, in each phenotypic group, there are more than 4 samples with zero counts. The noted genes have a relevance weight well above the threshold of \(\:\delta\:W=\:\) 0.01, and they show very weak correlation, leading to their survival as classifier genes. The sparse and very low gene count in a few samples only can be viewed as noise. Despite the noise, these genes survive as relevant to frailty and most of them would weakly correlate among them to further survive as classifier genes. To avoid the chancy scenario of genes with noisy counts surviving, it is best to take them off, based on a given criterion: genes with a total normalized count across the training samples that is less than a given fraction (here 20/22) of the number of training samples are removed. Prior to selection of relevant genes with the dataset of 22 training samples, any annotated genes with a total count less than 20 across all the samples are removed. This reduces the gene set from 20411 to 17222 genes. Classification of Samples (24 young and 18 frail) based on Genes that have survived noise removal The parameters that define the selection of relevant features are \(\:m=18\) , \(\:k=7\) , and \(\:\delta\:W\) = 0.03. It is worth noting that the surviving set of relevant genes and their weighted order of relevance depends on the number of nearest hits and misses, as well as the random samples chosen. This affects the eventual selection of classifier genes. It is worth considering making a series of runs in which the choice of random samples is different in each run, resulting in different sets of classifier genes. It would be worth confirming that the different sets of classifier genes are reasonably equivalent by producing similar patterns in classification. The most recurrent and most relevant genes across the different sets of classifier genes could become a final choice set. A number of independent runs in which different random samples are used to determine the relevant genes are discussed. The same selection criterion for non-correlation, \(\:R<\:{R}_{th}\) , where \(\:{R}_{th}\) = 0.5, is applied in all runs. After obtention of the classifier genes, the datasets of training and validation samples that are composed of the classifier genes only are submitted to k-means clustering, in order to verify how well the samples become classified into 2 separate clusters, based on the classifier genes only. It has been shown earlier in this work that the outcomes of k-means and kNN classification are quite similar and coincident in some datasets; k-means is chosen here for its relative simplicity in report of the results. The results in 5 cases ( A , B , C , D , E ) of the independent runs are summarized in Table 1 . The sample misfits shown do not cluster with the rest of their phenotype in all the k-means runs. If the sample does not cluster in a significant number of the runs (more than a quarter), it is also defined as a misfit. Such is the case for sample s517_F_Yo in case D. In each k-means run, the number of iterations required for convergence is in the range of ~ 2–5. The number of classifier genes obtained in the independent selection runs is in the range 9–12. A classification score for the classifier genes relative to the entire gene-set is above 90%, except in case A, where the score is 82.5%. Interesting finding: Based on these classifier genes, the few samples that do not regularly cluster with their phenotype are samples of female gender! This may be explained by the tendency to exercise less in females, resulting in their relative disposition to frailty. Table 1 Outcome of gene selection and k-means classification in 5 independent selection runs Case (# of relevant genes) # of classifier genes Classifier Genes Sample misfits score A (5791) 12 SEMA6A, FAM153A, CRADD-AS1, VWC2L, LRRC3B-AS1, CYP3A43, MIR5697, FAM178B, CFAP107, RPS2P32, NACA2, NUDT12 s301_M_Fr , s305_M_Fr , s316_F_Fr , s322_F_Fr , s507_F_Yo , s517_F_Yo , s518_M_Yo 82.5% B (6083) 10 C10orf71-AS1, RANBP17, CRADD-AS1, CDC26P1, RTL9, CYP3A43, MIR3682, RPS2P32, CFAP107, RNA5SP466 s516_F_Yo , s522_F_Yo 95% C (5858) 10 C10orf71-AS1, CRADD-ASI, CDC26P1, GSG1L, CYP3A43, LRRC3B-AS, RPS2P32, CXCL3, FAM238B, CFAP107 s516_F_Yo , s522_F_Yo 95% D (5697) 9 C10orf71-AS1, CRADD-AS1, CYP3A43, CDC26P1, GSG1L, RANBP17, RTL9, RPS2P32, CFAP107 s516_F_Yo , s517_F_Yo , s522_F_Yo 92.5% E (3861) 10 C10orf71-AS1, CDC26P1, RTL9, CRADD-AS1, LINC01121, CYP3A43, RANBP17, RPS2P32, SHISA8, RNA5SP466 s516_F_Yo , s522_F_Yo 95% Making Use of the determined Classifier Genes Once the classifier genes are determined, it is of interest to test the frailty condition for an arbitrary sample similar to the approach discussed in the classification of validation samples. Given that the rna-seq counts for the test sample will be obtained from an experimental set-up different from the set-up in which the rna-seq counts of the training samples were obtained, batch effects will have to be accounted for in order to determine the rightful Euclidean distance of the test sample from the training samples. The variation on the handling of biological samples in different rna-seq quantitation experiments influences the number of counts recorded for RNA transcripts. As we aim to test for the classifier genes in datasets that have been characterized for frailty in other experiments, with lack of knowledge of batch effects on the rna-seq counts in the separate experiments, we resort to clustering of the test samples based on the classifier genes. Separation of the test samples into their phenotypically characterized groups based on the classifier genes alone will be a strong confirmation of the authenticity of the classifier genes. Again, k-means clustering is used for this endeavor. Analysis of Test Dataset The classifier genes are tested across a GEO dataset of peripheral blood mononuclear cells (PBMC) genes in 16 middle aged individuals that have been tested for frailty. Although the tissue from which the genes of the test dataset are derived is not the same tissue from which the classifier genes are derived, it is of interest to reveal the tendency to identify frail samples from non-frail samples. The table of 39374 normalized and annotated RNA-seq gene counts data composed of 16 samples is read into RStudio 4.3. There are no duplicate gene names. A MDS plot of the 16 samples based on the original gene-set is displayed in Fig. 5 − 1 . There is no clear-cut division between distinct frail and non-frail regions, but there are regions of space which are dominated either by frail samples or by non-frail samples. Examples of identified regions can be named. Samples frail_1, frail_2, frail_4 are seen to form a contiguous region in which there is no intervening sample of the contrast phenotype. Other such regions are: samples non-frail_1, non-frail_2, non-frail_5; and samples non-frail_6, non-frail_7 and non-frail-8; and samples frail_3, frail_6, frail_7, frail_8. The results of k-means clustering, displayed in Fig. 5 − 2 , consistently show that the following groups of samples belong to one cluster indexed as \(\:1\) : frail_1, frail_2, frail_4; non-frail_1, non-frail_2, non-frail_3; and non-frail_5, non-frail_6, non-frail_8. All these samples are also located on the left side of the MDS plot, at leading logFC dim1 < 0, of Fig. 5 − 1 . The remainder of samples belong to the cluster indexed as \(\:2\) . For k-means to display results that provide observations similar to those noted from visualization of the MDS plot, a multi-cluster k-means approach should be adopted. The purpose is to identify clusters that show similar sample compositions as the noted regions in the MDS plot. The number of clusters adopted is four (4). Classification of Test Dataset Samples based on k-means with 4 clusters 4-centroid k-means with random initialization is applied to the entire gene-set of the test dataset. 5 out of 6 samples in cluster indexed as 4 are frail samples – frail_3, frail_5, frail_6, frail_7, frail_8; non-frail_4 is the only non-frail sample. In the cluster indexed as 3, there are 2 frail samples (frail-1 and frail-2) and 1 non-frail sample (non-frail-3). In cluster 2, there are 3 non-frail samples (non-frail_1, non-frail_5, and non-frail_6) and 1 frail sample (frail_4). And finally, cluster 1 has only non-frail samples – non-frail_2, non-frail_7, non-frail_8. [See Fig. 6-a for display of the samples in clusters]. The k-means clustering reflects the proximity of samples in the MDS plot, and most samples cluster with samples of their own phenotype. Samples of a minority phenotype in a cluster are called misfits , similar to the earlier description provided of samples that do not cluster with the rest of their phenotype. There are thus 3 misfits upon clustering with the entire gene-set, which are: non-frail_4, non-frail_3, and frail_4. k-means is then applied with 4 centroids to classify the test dataset samples based on classifier genes. The classification of samples closely matches that obtained with the entire gene-set, validating the adequacy of the classifier genes. In most scenarios, the 3 frail samples – frail_1, frail_2, and frail_4 – are classified in a cluster separate from the remaining frail samples, while the frail samples frail_6, frail_7, and frail_8 are always in one cluster, to closely depict what has been witnessed as sub-regions of a unique phenotype in MDS. The clustering meets the expectations of classification in which the clusters are having the majority of samples as one phenotype. k-means is performed with the different sets of classifier genes obtained in the previously described cases A - E . k-means Classification of Test Samples based on defined sets A-E of classifier genes The classifier genes obtained from a given dataset may not all be present in a test dataset with samples obtained from a different experiment. If the derived classifier gene is absent from the test dataset or has a zero count in all the test samples, it is replaced by the correlating gene that is closest by relevance to frailty in the training dataset. Among the classifier genes, the top gene C10orf71-AS1 is present but not expressed (zero count) across all samples in the test dataset. To enable the described replacement of genes, tables of the top five (5) correlating genes that are closest by relevance to the classifier genes are produced and used; see Tables 6 in the methods section. In case B, C10orf71-AS1 is replaced by “STRAD8”, CDC26P1 is absent and is thus replaced by “GSTZ1”, RNA5SP466 is absent and is thus replaced by “ANKLE1”; in case C, C10orf71-AS1 is replaced by “TEX56P”, CDC26P1 is replaced by SMIM10L1 ; in case D, C10orf71-AS1 is replaced by “UNC13C”, CDD26P1 is replaced by CD27-AS1; and in case E, C10orf71-AS1 is replaced by “OIP5-AS1”, CDC26P1 is replaced by C1orf21, RNA5SP466 is replaced by CFAP20. The outcomes of classification of the test samples based on the sets of classifier genes A-E are summarized in Table 2 . Diagrams of clusters containing the test samples are depicted in Fig. 6b and Fig. 7 a-d. The clustering outcomes are consistent within 4 independent runs, and in each run, convergence occurs in less than 5 iterations. Like in the evaluation for training and validation samples, the classification score is evaluated relative to the outcome of classification that is based on the entire gene-set, such that only dissimilar sample misfits in both outcomes account for the score. The number of dissimilar misfits to the classifier set outcome and the entire gene-set outcome is indicated in the 3rd column of Table 2 . Classification scores are \(\:\ge\:\) 80% in all 5 cases, with the lowest at 80%, and the highest at 93.75%. In all cases, highlighting samples that are recurrently misfits, the sample non-frail_4 is a misfit in all 5 cases, sample non-frail_5 is a misfit in 3 cases, sample non-frail_3 is a misfit in 2 cases, and sample non-frail_8 is a misfit in 2 cases. All the noted samples, except non-frail_8, are the only misfits in the classification with the entire gene-set, as this shows how the classifier genes closely represent the entire gene-set in classification. Notably, sample non-frail_4 can be seen in the MDS plot of Fig. 5 − 1 to be significantly closer to frail samples than to non-frail samples. The k-means results can be qualified as robust! Table 2 Outcome of classification of test dataset samples. Case Sample misfits # of dissimilar misfits score A non-frail_4, frail_4, frail_5, non-frail_3, non-frail_5 2 87.5% B non-frail_4, non-frail_1, non-frail_8 2 87.5% C non-frail_4, non-frail_2 1 93.75% D non-frail_4, non-frail_8, non-frail_3, non-frail_5 2 87.5% E non-frail_4, non-frail_8, non-frail_5, frail_5, frail_1 4 80% Discussion of Top Frailty Genes A survey is done to see how the relevant frailty genes compare to notable frailty (or aging) genes in the literature. Notable aging genes are drawn from the literature for comparison with the genes that are relevant to frailty in this work. Investigations conducted in Ref. [ ] resulted in the identification of top 20 genes that are upregulated with aging, and top 20 genes that are downregulated with aging. The study was done with 53 healthy individuals (22–83 years old) on 57,205 protein-coding and non-coding RNAs. The healthy condition of the individuals provides an interpretation of gene expression results that is free from the impact of disease. The RNA expressions of the identified genes linearly increase (for upregulated genes) or decrease (for downregulated genes) with age in linear regression plots of the cited Ref. The ranks of the identified notable genes among genes relevant to frailty in this work are determined. The gene selection outcome from the dataset of samples that survived outlier removal by k-means clustering with relevance threshold \(\:\delta\:W\) = 0.01 is used as a reference for sourcing the relevance rank. The rank and weight of every notable gene is recorded in Table 3 : the genes that are upregulated with age are tabulated in Table 3 − 1 , and those that are downregulated with age are tabulated in Table 3 − 2 . Out of the top 20 upregulated genes, one of them is without a gene name, leaving 19 genes in Table 3 − 1 . Out of the top 20 downregulated genes, 13 of the genes have gene names, as that is the number of genes in Table 3 − 2 . Genes with “NA” in the rank and weight are those that are not found among the relevant genes – there are six such genes: TET2, EFEMP1, FAM49A, SMIM2, MYLPF, and MINOS1. Only TET2 and EFEMP1 are among the annotated genes of the training dataset, as their weight of relevance to frailty is below the accepted threshold of 0.01. The remaining four genes are completely absent from the training dataset of annotated genes. Among the 32 notable aging genes tabulated, 11 genes are among the top 10% rank of relevant frailty genes, as this seems a remarkable representation of the notable genes among the genes that have been determined to be relevant to frailty. Looking at any set of classifier genes, it may be difficult to identify among them notable frailty genes from the literature. It suffices to show, however, that the classifier genes correlate with the identified notable genes. Notable aging genes that correlate with the top 12 classifier genes are found, enabling thus to witness the link between the classifier genes and the notable aging genes. This correlation is tabulated in Table 4 . As a remarkable feat, 26 of the 32 notable aging genes of Table 3 − 1 correlate with the 12 classifier genes, even if none of the notable genes is among the given classifier set. It can also be observed that the more highly ranked classifier genes correlate with a bigger number of notable frailty genes, as 7 notable frailty genes correlate with the top classifier. The classifier genes that correlate with several notable genes may not have a fully or appropriately defined role in frailty in the literature. This insufficiency provides grounds for determining or investigating the roles of such classifier genes! The top 500 relevant genes are submitted to the Molecular Signatures Database (MSigDB) – a multicollection of databases among which are curated gene sets and databases of ontology gene sets, in order to find gene sets that overlap with the submitted genes. The online search for the 20 most populated gene sets is done for the 5 variants of relevance gene sets A-E, and the overlapping genes for every gene set are described. The results showing the gene-set names, the overlapping genes and their description are provided in files with the ‘tsv’ extension in the supplementary section. The files are named as overlap_A.tsv, overlap_B.tsv, overlap_C.tsv, overlap_D.tsv, and overlap_E.tsv, for the 5 different variants A-E, respectively. Although searches based on the different variants display different combinations of gene sets, there are gene sets that are recurrent across the different combinations. Such gene sets and some of their constituent genes are tabulated in Table 5 . In Table 5 , are also indicated gene sets that are of characteristic evidence to frailty, e.g. the gait disturbance gene set. The gene-set names in Table 5 are indicative of processes and conditions at different levels of organization. Parts of some gene set names at the cellular level are: mitochondrion, apoptosis, target genes, protein modification, nuclear complex, and microRNA. Parts of gene set names at the tissue and organ level are: skeletal muscle cells, fetal lung cells, and pancreas ductal cells. Some of the conditions indicated by the gene set names are: Alzheimer’s disease, nasopharyngeal carcinoma, bladder cancer, chronic myelogenous leukemia, and gait disturbance. A metabolic process that is indicated is phosphorus metabolic process. Table 3 − 1 . The rank and weight of notable [upregulated with age] aging genes in Ref. [15] among genes that are relevant to frailty. gene rank \(\:W\) Functions and characteristics, described in Ref. [15] C12orf75 125 0.213865 Energy metabolism, insulin signaling, skeletal muscle weakness CDKN2B 85 0.226386 p15 tumor suppression, aneurysm formation, aging SKAP2 7 0.322864 Sarcomere function and regulation CFAP61 72 0.233005 High expression in old skeletal muscle, skeletal muscle weakness PLAG1 96 0.221592 Bovine fetal development and body size LGI1 119 0.214702 Skeletal muscle weakness, limbic encephalitis, and epilepsy IRS2 701 0.13417 Insulin signaling, lipid metabolism in skeletal muscle, mice longevity FEZ2 233 0.188437 fasciculation, axonal elongation, skeletal muscle aging PFKFB2 412 0.159812 Heart isozyme ESRRG 7844 0.01944 Type I muscle fiber development, estrogen signaling, obesity susceptibility FAM83B 1737 0.09142 Hypoxia response pathway, aging, skeletal muscle weakness FAM117B 8468 0.015853 Sarcoidosis susceptibility, candidate for early onset myocardial infarction CRIM1 3191 0.061272 Skeletal muscle aging, smooth muscle contractility, CNS development NR2F2 2918 0.065517 Myogenesis, skeletal muscle development, aging, muscular dystrophy LPP 4551 0.044722 Smooth muscle differentiation, cell migration and proliferation TET2 NA NA KCNQ5 1847 0.08812 Potassium channel subunit EFEMP1 NA NA ENAM 8813 0.01402 Tooth enamel formation Table 3 − 2 . The rank and weight of notable [downregulated with age] aging genes in Ref. [15] among genes that are relevant to frailty. gene Rank \(\:W\) Functions and characteristics, described in Ref. [15] MYLK4 344 0.16906 weakness, contraction, motility, TGF-beta pathway, circadian rhythms MYL1 3996 0.050828 Expressed in fast-type muscle LDHA 4925 0.040896 Porcine energy regulation, glycogen metabolism, and glycolysis of skeletal muscle FAM49A NA NA TPM1 1516 0.098513 Modulates actin-myosin interactions TPI1 712 0.133497 Glycolysis enzyme, energy generation for muscle cells in chickens SLN 1665 0.093632 Inhibits rat sarcoplasmic Ca2+-ATPases, thermogenesis, muscle performance TNNT3 2938 0.065257 Skeletal muscle contractibility, porcine muscle development, tropomyosin binding SMIM2 NA NA TUBA8 8532 0.01553 Intracellular transport of macromolecules and cytoskeletal support MYLPF NA NA MINOS1 NA NA MSS51 3413 0.058304 Skeletal muscle-specific modulator of cellular metabolism Table 4 Record of the notable aging genes from Ref. [15] that correlate with a set of 12 frailty classifier genes. Rank Classifer Gene Correlating notable aging genes in Ref. [15] 1 C10orf71-AS1 C12orf75, CDKN2B, SKAP2, CFAP61, PLAG1, LG1, FEZ2 188 CDC26P1 IRS2, PFKFB2, MYLK4, TPI1 704 CRADD-AS1 None 948 RANBP17 TPM1, SLN 1544 CYP3A43 None 1551 LINC01121 FAM83B, CRIM1, NR2F2, KCNQ5 2757 RTL9 MYL1, TNNT3, MSS51 3190 SHISA8 LPP, LDHA, 5232 RPS2P32 None 5709 RNA5SP466 TUBA8, 6313 CCDC144NL-AS1 ESRRG, FAM117B, 7312 YBX2P2 ENAM Table 5 Summary Results of MSigDB query of gene sets that overlap with top 500 relevant frailty genes. The first word of every gene set name is the name of the database collection or the source of provision of the gene set. MSigDB gene set Description of gene set # of overlap genes Example genes GRAESSMAN APOPTOSIS BY DOXORUBUCIN DN Genes downregulated in ME-A cells (breast cancer) undergoing apopotosis in response to doxorubicin 65 SCL44A2, IRF1, FEZ2, URI1, PP1CC, HSPE1, SFXN4, STX17, RFK, SIRT4, KIF5B, CTNNBIP1, HNRNPL, NAGK, GJC1, STRADA, PLP2, ZNF768, TRIM44, PPP1CC, SRSF1, TARS2 BLALOCK ALZHEIMERS DISEASE UP Genes up-regulated in brain from patients with Alzheimer’s disease 59 SEMA6A, FEZ2, USP36, STX4, ACIN1, NNT, GBE1, PKRCH, STRADB, NCKAP1, ITGA6, ITGA1, TACC1, RGCC, SIRT4, URI1, RFK, CRADD, IGF1, HNRNPK, ERBB2, TRAP1, RTL10, PARP1, CYCS, DAPK3, TNFRS10B, SLIT2, CDKN1A, ADAM8 GOCC MITOCHONDRION A semiautonomous, self-replicating organelle that occurs in varying numbers, shapes, and sizes in the cytoplasm of eukaryotic cells. It is notably the site of tissue respiration 67 PDP1, NKTR, GRSF1, PDHX, MRPL38, DHX32, PARP1, FAM210A, MT-ND3, RTL10, SMIM20, URI1, MT-CO2, NNT, MTHFDIL, CHCD3, MFF, MCM7, GOT2, TARS2 PAX3 TARGET GENES Genes containing one or more binding sites for Uniprot:P23760 (PAX3) in their promoter regions (TSS-1000,+100 BP) as identified by GTRD version 20.6 ChIP-seq harmonization. 70 TEX56P, SCYL2, PCDH17, SEMA6A, HNRNPC, TNFRSF10B, MAP4K, ITFG2, SKAP2, UBE21, ITGA6, CTNNA3, TMEM19, FEZ2 MIR561_3P Genes predicted to be targets of miRbase v22 microRNA has-miR-561-3p in miRDB v6.0 with MirTarget v4 prediction scores > 80 (high confidence targets) 33 RANBP3, SLC25A15, PCDH9, C1orf21, POPDC3, CFAP61, URI1, STX17, PPA2, POU2F1, DAZAP1, SPRED1, ZBTB18, HAPSTR1, PHTF2 DODD NASOPHARYNGEAL CARCINOMA UP Genes up-regulated in nasopharyngeal carcinoma (NPC) compared to the normal tissue 58 SEMA6A, CBX1, SRSF7, STRAP, MCM3, TIPRL, ANKIB1, SLC16A1, UBE2Z, ITGA1, MAP2K4, NOLC1, KLHL5, PAIP1, SLC16A1, EIF3J, TIMMDC1, TRMT61B, SCYL2, NNT, MFF, SRSF1 NAKAYA PBMC FLUMIST AGE 18 50YO 7DY UP Genes up-regulated in peripheral blood mononuclear cells 7d vs 0d in adults (18–50) after exposure to FluMist, time point 7D. 61 GRSF1, SFXN4, MTIF2, ATPAF1, FEZ2, DYM, RPRD1A, EWSR1, GLG1, PPP2CB, COPS5, PPIE, ATP13A1, WRAP73, CEP164, MORC2, RNF170, ARSG, SCOC, TERF2, NNT, CHCHD3, SRSF1, GOT2 HP ABNORMAL FUNDUS MORPHOLOGY Abnormal fundus morphology 55 CDKN2B, SEMA3E, LPAR6, ANKRD11, PEX2, RNF170, DTNBP1, TENM3, SCYL2, COA8, MT-ND4, PDHX, TIMMDC1, SCL25A46, MT-CO2 DIAZ CHRONIC MYELOGENOUS LEUKEMIA UP Genes up-regulated in CD34 + cells isolated from bone marrow of CML (chronic myelogenous leukemia) patients, compared to those from normal donors. 56 TRAK2, NNT, CYCS, MTIF2, PPA2, NIPSNAP2, SCP2, HNRNPC, KPNA4, MAP4K4, PTP4A2, NDEL1, GBE1, PARP1, SLC16A, PAIP1, CCNG1, PLP2, CDC42BPA, VTI1B, PEX2, SART3 THAKAR PBMC INACTIVATED INFLUENZA AGE 21 30YO RESPONDERS 28DY DN Genes down-regulated in peripheral blood mononuclear cells 28d vs 0d in young adults (21–30) after exposure to inactivated influenza vaccine, time point 28D 44 STRADA, STX4, USP36, NISCH, NAGK, RBM23, SEC24C, MAN2C1, WDR6, ADAM8, ARFGAP2, RANBP3, ZFHX3, CEBPB, ATP13A1, VPS52, ITFG2, NKTR TFEB TARGET GENES Genes containing one or more binding sites for UniProt:P19484 (TFEB) in their promoter regions (TSS-1000,+100 BP) as identified by GTRD version 20.06 ChIP-seq harmonization. 56 MFF, LIG1, FA210A, SKAP2, PRKCH, PRRC2C, MAPKAP1, CEP164, KHDRBS1, STRADA, STX4, AC1N1, HNRNPL, PHFT2, HAPSTR1, ANKRD11, SLC22A18, PPP1R3E, GABBR2, DNAH11, STRADB, MFF RUBENSTEIN SKELETAL MUSCLE ENDOTHELIAL CELLS Same as name of gene set 20 NKTR, ITGA6, PTPRB, PHACTR2, F8, AQP1, ITGA1, TACC1, RGCC, FLT1, SPTBN1, MGLL, MYCT1, NEAT1 HE LIM SUN FETAL LUNG C5 LARGE PRE B CB CELL Large pre-B 53 PCDH17, CLK2, UBXN2A, C12orf75, DDB2, SART3, TELO2, EDC4, RBMX2, CCNG1, DADK3, NFATC2IP, MCM7, GUSB, NDUFC1, PPA2, MCUR1, SCCPDH, MRPL13, TRAP1, MTIF2, PDHX, CTNBIP1, ADK, CBX1, DAZAP1, SFXN4, MRPL15, ENO3, GOT2, MLEC GOBP APOPTOTIC PROCESS A programmed cell death process which begins when a cell receives an internal (e.g. DNA damage) or external signal (e.g. extracellular death ligand), and proceeds through a series of biochemical events (signaling pathway phase) which trigger an execution phase. 65 SEMA6A, PLAG1, CDKN1A, TMEM163, SLC16A1, IGSF3, ADAMTSL4, DPF2, ERBB3, GFOD1, PCSK7, TJAP1, FCRLA, HNRNPC, PPFIBP1, KCNMA1, KLF5, EIF4H, MRPL13, MTIF2, DCLK1, SPRED1, H1-0, HSD3B7, IGF1, ZFHX3, SASH1, SLC16A1, PRCKH, LDLRAD4, C16orf74, STX4, USP36, CDH5, PDP1, LDLRAD4, MAPK6, RIN1, NDUFC1, GREM1, DPF2, RNF152, IP6K2, ERBB3 GOBP REGULATION OF PROTEIN MODIFICATION PROCESS Any process that regulates the frequency, rate or extent of the covalent alteration of one or more amino acid residues within a protein 50 NNT, CDKN2C, CCNG1, SVIP, UBXN2A, DCUN1D5, STRADB, CDH5, MAPKAP1, STRADA, CDKN1A, IP6K2, GREM1, TNFRS10B, IL6R, EFNA1, ITGA1, NCK1, RGN, UCN, PHIP, PRNP, ERBB2, CCAR2, SLIT2, ACE, HDAC4, TAB2, NELFA, DOCK7, SPTBN4, GPS2, SASH1, ATP2B4, PARD3, TRIM44, NNMT SKIL TARGET GENES Genes containing one or more binding sites for UniProt:P12757 (SKIL) in their promoter regions (TSS-1000,+100 BP) as identified by GTRD version 20.06 ChIP-seq harmonization. 68 SEMA6A, CDKN2C, CMC2, TNFRS10B, DCAF5, TARS2, HMBOX1, H1-10, EWSR1, PNPLA7, DDX51, ZXDC, PYGO2, ATP13A1, FUS, HSPE1, CDC42BPA, AKAPBL, RPS27L, CCAR2, CYCS, CLN8, QRICH1, COA8, HECW2, NEAT1, WDR6, RIN1, LDLRAD4, MTHFDIL, PRKCH, MLEC, MCM7, MRPL13, MTIF2, NUP85 MURARO PANCREAS DUCTAL CELL Same as gene set name 52 NNT, CBR1, TNFRS10B, ITGA6, ATP2B4, PARD3, DERL1, CDKN2B, SLC44A2, HEG1, GNG12, GALNT1, H3-3B, PPFIBP1, IGSF3, KLF5, PPIC, IFNARI1, KDM5B, KLHL5, PDP1, MAPK6, TRAK2, MAPKAP1, CAST GOBP PHOSPHORUS METABOLIC PROCESS The chemical reactions and pathways involving the nonmetallic element phosphorus or compounds that contain phosphorus 66 CDC42BPA, SASH1, PPP1CC, URI1, STRADA, DCLK1, PHIP, SCL44A2, NAGK, ERBB2, LCLAT1, PDP1, PRCKH, TNFRSF10B, MAPK6, SEMA6A, PARP1, DRC3, COA8, DDX51, ATP13A1, TRAK2, PHACTR2 OSMAN BLADDER CANCER Genes upregulated in blood samples from blood cancer 27 EWSR1, GALNT1, NNT, TIPRL, ATF71P, APPL1, KDHRBS1, NRIP1, HNRNPL, PTP4A2, TRAK2, NKTR HOUNKPE HOUSEKEEPING GENES List of 1130 human and mouse housekeeping genes. This list is the overlap of human genes stably expressed across 52 tissues and cell types. 45 TRMT112, QRICH1, HSPE1, DCAF5, PPP1CC, STX4, KHDRBS1, CBX1, VTI1B, THRAP3, EIF4H, UBE21, PARP1, MLEC, SART3, CCAR2, DDX51, VPS52, NXF1, KPNA4, SCYL2, HAPSTR1, NTSC2, LDB1, PPP2CB, MMADHC, NFAT2CIP, ATPAF1, RBMX2, PITHD1, H3-3B, DERL1, APPL1, STRAP, ERLEC1, PBDC1 GOCC NUCLEAR PROTEIN CONTAINING COMPLEX A stable assembly of two or more macromolecules, i.e. proteins, nucleic acids, carbohydrates or lipids, in which at least one component is a protein and the constituent parts function together in the nucleus 50 NUPB5, E2F3, RBMX2, MCM7, EXOSC7, CTNNBIP1, MCM5, SART3, INTSI3, MCM3, PPIE, WDR6, HDAC4, NELFA, PYGO2, DOCK7, INO80E, ELOA, CDKN1A, HNRNPC, DPF2, TEAD1, CXXC1, HNRNPM, CBX4, PPP2R1A, DMAP1, NOC2L, KMT2E, NKF1, SMARCD2, INTS5, TERF2, TDRD3, KMT2B, CRY2, ACTR5, MVP HP GAIT DISTURBANCE Gait disturbance 50 GBE1, RNF168, PDHX, TELO2, PDP1, WDRB, STRADA, SLC25A46, MORC2, MFF, COA8, CP, FUS, KDM5B, NTSC2 MATERIALS AND METHODS Processing of GEO datasets The genes in the input dataset are annotated as gene IDs, with 2 digits after a dot in the ID. The two digits are meant to indicate the version of the annotation technology used, and thus do not impart any biological meaning. The 2 digits and the dot are trimmed off from the gene IDs. Any rows in the data that are bearing duplicate gene IDs are also deleted from the dataset, resulting in 58238 genes from an initial number of 58283. The dataset that is used to construct the set of classifier genes is raw counts of RNA-seq data obtained from muscle biopsies of vastus lateralis, deposited in the Gene Expression Omnibus (GEO) database under the title “Muscle function in fit and (pre-)frail males and females”, made public in Feb 04, 2023, and accessible from the Gene Expression Omnibus (GEO) with accession number GSE144304. The RNA-seq counts were obtained with the Illumina NextSeq 500 (Homo sapiens) platform from samples that were submitted in Jan 27, 2020. The total number of samples in the dataset is 80, with each sample coming from a separate individual. The individuals are phenotypically categorized as young, fit, or frail. Old subjects (> 75 years) were screened for frailty using the Fried criteria. (e.g. weight loss, exhaustion, low physical activity, slowness and weakness) [ ]. Based on these criteria the old subjects were divided into two groups; fit (no frailty indicators present) and frail (≥ 1 frailty indicator present). The young subjects were individuals 20–30 years of age with a healthy BMI (20–25 kg/m 2 ). Several exclusion criteria were applied to the young subjects including cardiac failure, COPD, anemia, dementia, cancer, neuromuscular disorders and diabetes mellitus type I and II. Detection of Outlier Gene Counts and Normalization . Normalization of genes and gene count outlier detection and replacement is achieved by programming routine functions in the R package DESeq2 . The function DESeqDataSetFromMatrix() is applied on the raw counts data, with the samples of the same phenotypic condition noted in one distinct group, to produce an object named ‘dds’. All genes with the sum of counts across all samples less than 10 are removed by applying the function rowSums(count(dds)) > = 10. The function DESeq is further applied on ‘dds’ as DESeq(dds) to obtain an object named ddsDE. DESeq() by default detects and replaces outlier counts with replacement counts when the number of biological sample replicates is greater than or equal to seven (7). For conditions that contain seven or more replicates, DESeq2 replaces the outlier counts with an imputed value, which is the trimmed mean over all samples, scaled by the size factor; and then re-estimates the dispersion, LFCs and P values for these genes. For a given sample, the median of the ratios of counts to the geometric mean for each gene across all samples is the size factor for that sample. The automatic identification and replacement of outliers, however, leaves the analysis blind to the detection of gene outliers. A procedure to identify the outlier genes can be implemented to allow for removal of the genes. In the ddsDE object, there is a matrix of preserved original counts identified as assays(ddsDE) $ counts and a matrix of replacement counts – assays(ddsDE) $ replaceCounts that has a modified count for outlier counts only. The outlier counts are identified as counts that do not equal one another during comparison of counts in the two matrices. Some genes are outliers in multiple samples for the GSE144304 dataset, identifying a total of 60 outlier counts. One (1) gene is an outlier in 3 differing samples, and seven (7) genes are outliers in two (2) differing samples. The number of unique genes with outlier counts is thus fifty-one (51). Those genes that were identified as outliers only had outlier counts in very few of the samples of either the young or frail category. Any sample is only having few outlier gene counts, less than 5 out of a total of about 50 000 counts. As a result, it is deemed reasonable to strip off the gene or modify the outlier count, instead of removing the sample that has it as an outlier. Here the outlier counts are replaced by the recently described procedure. The counts are normalized from ddsDE through the function counts (ddsDE, normalized = TRUE). DESeq2 performs an internal normalization where the geometric mean is calculated for each gene across all samples. For a given sample, the median of the ratios of counts to the described geometric mean is the size factor for that sample. This procedure corrects for library size and RNA composition bias, which can arise for example when only a small number of genes are very highly expressed in one experimental condition but not in the other. A description of the procedure of removal of outliers and normalization of counts is properly documented in Ref. [ ]. Identification of Outliers based on MDS plots The MDS plot for the original dataset of GSE144304 shows that samples s326_M_Fr and s525_F_Yo are outstandingly distant from the remainder of samples (see Fig. 7 ). These two distant samples are taken off, and the MDS plotted again; see Fig. 8 . In the new MDS plot of 48 samples, sample s514_F_Yo seems significantly distant from the remainder of young samples, and seems closer to frail samples than to young samples. Sample s323_M_Fr seems clustered by young samples. Samples s508_M_Yo and s530_F_Yo seem closer to more frail samples than young samples. The named samples – s508_M_Yo, s530_F_Yo, and s514_F_Yo, and s323_M_Fr – are deemed outliers, and are removed, leaving 22 samples in each category of frail or young samples, out of the initial 24 frail samples and 26 young samples. As an alternative procedure to demonstrate that there are samples that are either extremely distant from the remainder of samples of their phenotype in the GSE144304 original dataset of 50 samples, or are closer to samples of the contrast phenotype, the density of samples around each sample can be computed to show the number of samples that are within a given Euclidean distance from that sample. This defined distance or radius is defined to be three times the mean value of the matrix values of Euclidean distances between all the samples. The region within the defined radius of a sample can be called the neighborhood of the sample. The density of samples in the neighborhood of each sample is plotted and displayed in Fig. 9 . The number of samples within the neighborhood of each sample is 47, except for samples s326_M_Fr and s_525_F_Yo that have only 1 sample in their neighborhood. And the neighbor of any of these exclusive samples is the other of the exclusive samples, meaning that they are closer to one another than to the rest of all the samples. With the noted extreme outliers removed, the density of samples in the neighborhood of each sample is plotted and displayed in Fig. 10 : in this plot, the radius that defines the neighborhood of a sample is the mean value of the matrix of Euclidean distances for all the remaining 48 samples. Samples s314_F_Fr and s514_F_Yo have the least number of neighbors – six (6), as they correspond to sample indices 8 and 35, respectively. However, for sample s514_F_Yo , all the six neighbors are frail samples, namely: " s301_M_Fr ", " s314_F_Fr ", " s316_F_Fr ", " s320_F_Fr ", " s325_F_Fr ", and " s331_F_Fr ". And for sample s314_F_Fr , its six neighbors are mostly frail samples, with the exception of s514_F_Yo in that fold; the frail samples are s301_M_Fr , s305_M_Fr , s320_F_Fr , s325_F_Fr , and s331_F_Fr . Sample s514_F_Yo is not only relatively distant from the remainder of samples, but is having only frail samples in its neighborhood. It is thus an outlier. The results of this neighborhood analysis confirm the observations that have been noted from the MDS plots. The described procedure by which the samples density around each sample is computed to show the number of samples that are within a given Euclidean distance from that sample is useful for spotting samples that are evidently outliers. k-means clustering By the experience of several investigators on k-means clustering, the initial choice of centroids can determine the outcome of classification. A random initialization procedure is adopted here: half of the samples are randomly selected to belong to one cluster, while the other half belongs to another cluster. The centroid of each initial cluster is determined as an imaginary (or virtual) sample of all the genes, with each gene having a count that is the mean of the counts for that gene across samples belonging to that cluster. After the centroids are generated, all the samples are classified according to their closest Euclidean distance to the centroids. A sample is adopted into a cluster if the Euclidean distance of the sample to the cluster centroid is closest among all cluster centroids. The resultant cluster is having samples that are different from those initially chosen at random. From the new clustering of samples, new centroids are again generated. The process of generating new clusters, and then, new centroids, is repeated until the new clusters that are generated are completely the same as in the preceding iteration – this final stage of the iterative process is called convergence. The number of iterations leading to convergence of the k-means algorithm can be recorded. Removal of Potential Outliers using k-means A small experiment is run to show the importance of using a MDS plot to remove samples that evidently seem distant from the rest of samples of their phenotype, or samples that clearly mingle within the contrast phenotype. It is shown that the described samples can significantly shift the centroid of one cluster, leading to several samples not being categorized in their parent group. In order to see the effect of samples that were clearly seen as outliers in MDS plots on k-means clustering, k-means is performed in presence of those samples apart from samples s326_M_Fr and s525_F_Yo that were outstandingly distant. The k-means clustering that involves generating two initial centroids from a random selection of the remaining 48 samples consistently shows that all phenotypically young samples belong to the same cluster, except s514_F_Yo that always belongs to the same cluster as frail samples. Another noticeable outcome is that while most phenotypically frail samples also belong to the same cluster, 9 samples, which are s323_M_Fr, s315_M_Fr, s317_M_Fr, s318_F_Fr, s329_F_Fr, s302_M_Fr, s303_M_Fr, s304_M_Fr, and s306_M_Fr consistently show a belonging to the cluster having young samples. The sample s514_F_Yo which is the only phenotypically young sample that is clustered among samples that are phenotypically young lies significantly distant from the rest of frail samples in the MDS plot. This sample may be influencing the centroid of young samples by shifting it towards the frail samples, explaining why there is a consistent number of frail samples in the young group. The number of iterations in the clustering is about 5 to 10. The noted consistency is true within 12 consecutive runs of the k-means clustering. Taking off s514_F_Yo leads to altered results in the clustering. After removing s514_F_Yo from the dataset, and conducting the clustering iterative process in 12 independent runs, all phenotypically young samples belong to one and the same cluster. Frail samples that consistently intervene with the cluster of phenotypically young samples during all the 12 k-means runs are now fewer (5): s323_M_Fr, s318_F_Fr, s302_M_Fr, s303_M_Fr, and s306_M_Fr. And samples s315_M_Fr and s317_M_Fr intervene with the cluster of young samples only in 1 run out of the 12 runs. Among these 5 samples, s323_M_Fr was determined to be an outlier from observations of the MDS plots; samples s318_F_Fr and s303_M_Fr seem to be on the borderlines between frail and young samples of the MDS plot in Fig. 2 . The noticed alteration in results shows how an outlying sample can influence the clustering of samples in k-means. Removing the frail samples identified above in k-means as intervening, and also removing s514_F_Yo, leaves 18 frail samples and 24 young samples. This surviving dataset is again submitted to k-means to identify samples s508_M_Yo and s530_F_Yo as intervening with the cluster of frail samples: in 9 out of 12 runs for s508_M_Yo, and in 7 out of 12 runs for s530_F_Yo. All the remaining samples cluster within their phenotypic group in all the 12 runs. Samples s508_M_Yo and s530_F_Yo were determined to be outliers from visualization of the MDS plot in Fig. 2 . At this stage, all the samples that were determined as outliers from visualization of MDS plots have also been determined to be intervening in k-means. Selection of Relevant Genes in ReliefF ReliefF is the best known and most utilized Relief-based algorithm that uses \(\:k\) nearest hits and \(\:k\) nearest misses in the weight update for each instance of the training samples, weights the instances equally, and takes the average over the number of instances used [5,10]. Relief-based algorithms are independent of the induction algorithm, i.e. after selection of a feature subset, any suitable classification scheme can be employed. This relieves of the need for the computational intensity and complexity that is involved in having to iteratively generate a feature subset until the classification seems satisfactory as in wrapper methods and embedded methods that are induction algorithm dependent. For example, there are wrappers for Naïve Bayes and Support Vector Machine classification algorithms; and in embedded methods such as decision-tree based algorithms, feature selection is part of the classification algorithm. Only genes above a threshold weight \(\:\delta\:W\) are selected. A random number of samples chosen from training samples is used to perform selection, and for each sample, there is an equal fixed number of nearest samples hits and misses to be considered. In the ReliefF algorithm, genes are deemed relevant to providing a contrast between two phenotypic conditions by exercising a weight that is above the threshold weight \(\:\delta\:W\) . A randomly chosen number of samples is used to determine the weights for each gene. For each gene, the weight is assessed as a cumulative sum over the randomly chosen samples. For each chosen sample, the weight is calculated for every nearest-neighbor sample hit or miss of the sample, and is then summed over the nearest neighbors. In the summation, the weight is assigned a negative value for every hit, and a positive value for every miss. A sample hit is a sample of the same phenotype as the chosen sample, while a sample miss is one of contrasting phenotype. The gene weight between the chosen sample and any other sample is calculated as the absolute value of the difference between the gene counts, divided by a constant value for that gene. That constant value is the difference between the highest count and the smallest count for the gene over all the training samples. The number of nearest hits and misses used to determine the weights is generally set here at 7 or above for the large number of training samples per phenotype. Initially, the weight for each gene is zero (0), as it accumulates as a sum over the randomly chosen samples, and for each sample, it accumulates over the nearest neighbor hits and misses. The sum is then averaged over the product of the number of randomly chosen samples and the number of nearest hits or misses. This sum can be summarized as \(\:{W}_{n}\) in Eq. (2) . It is only different from the sum depicted in Ref. [5] in the sense that there is only one class of contrasting phenotypes. $$\:{W}_{n}=\:{W}_{n}-\:\sum\:_{j=1}^{k}\frac{\text{d}\text{i}\text{f}\text{f}\left(n,{X}_{i},{H}_{j}\right)}{k}\:+\sum\:_{j=1}^{k}\frac{\text{d}\text{i}\text{f}\text{f}\left(n,{X}_{i},{M}_{j}\right)}{k}\:\:\:\text{f}\text{o}\text{r}\:i\:\in\:\left[1,m\right]\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(2\right)$$ In Eq. (2) , \(\:n\) is the index for a gene, and runs from 1 to \(\:N\) , where \(\:N\) is the number of genes. \(\:i\) is the index for the random chosen samples, and runs from 1 to \(\:m,\) the number of the randomly chosen samples. \(\:{X}_{i}\) is the count for gene \(\:n\) in the random sample of index \(\:i\) . \(\:j\) is the index for the nearest sample hit or miss, and ranges from 1 to \(\:k\) – the number of nearest hits or misses. \(\:{H}_{j}\) is the count for gene \(\:n\) in the sample hit of index \(\:j\) , and \(\:{M}_{j}\) is the count for gene \(\:n\) in the sample miss of index \(\:j\) . The function \(\:\text{d}\text{i}\text{f}\text{f}\left(n,X,Y\right)\) for any two samples is defined in Eq. (3) . \(\:\text{max}\left(n\right)\) and \(\:\text{m}\text{i}\text{n}\left(n\right)\) are the maximum and minimum values of the counts of gene \(\:n\) across the training samples; as such, \(\:\text{d}\text{i}\text{f}\text{f}\left(n,X,Y\right)\:\le\:1\) . $$\:\text{d}\text{i}\text{f}\text{f}\left(n,X,Y\right)=\:\frac{\left|X\left(n\right)-Y\left(n\right)\right|}{\text{max}\left(n\right)-\text{m}\text{i}\text{n}\left(n\right)}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(3\right)$$ Note that the division by \(\:k\) in Eq. (2) occurs \(\:m\times\:k\) times, and, thus setting a limit to the value of \(\:{W}_{n}\) : \(\:{W}_{n}\:\le\:1\) . The random samples are chosen by first generating a random number that is of equal distribution between 0 and 1. The generated number is then linearly projected to a number that is in the range [1, m ], where m is the number of random samples to use. To determine the nearest hits and misses for a sample, the Euclidean distance is determined between the sample and the remainder of the samples. The Euclidean distance accounts for all genes in the dataset as independent features. The samples of either phenotype with the \(\:k\) shortest Euclidean distances from the random sample of interest become the \(\:k\) nearest sample hits and sample misses. The Euclidean distances between samples are computed a priori and stored in a table. The table is a matrix of elements which are the Euclidean distances between samples that are the corresponding row and column names of the matrix. The distance between a sample and itself is zero (0), as this makes the diagonal elements of the matrix all equal to zero. Gene Correlation Before implementing the R code that determines gene correlation, the relevant genes are ranked in order of relevance. A null vector that will contain the classifier genes is initially defined in the R code. It is null, because it has no elements initially, but will increase in size as non-correlating genes are found and appended to it. The R code is written and run to produce the classifier genes based on the condition that all genes that correlate with the topmost relevant gene based on the criterion \(\:R>\:{R}_{th}\) are eliminated from the set of relevant genes in the first (1st ) iteration. After the 1st iteration, the topmost relevant gene is appended to the null vector as its first element, and the gene is also discarded from the running set of relevant genes. This leaves the correlating gene that is closest by relevance to the topmost relevant gene as the top in the list of running relevant genes. The iterative process is repeated until there is nothing left in the running set of relevant genes, as all the non-correlating genes have migrated to the vector of classifier genes. Besides producing the classifier genes, the top 5 genes that are closest in relevance to each classifier gene are produced. This is possible because the indices of the classifier genes in the vector of the original relevant genes are stored during the iterative correlation process. These indices are references to producing the mentioned top 5 genes. The top 5 supplementary genes are potential replacements for the specific classifier gene in case it is absent from the test dataset. A Table of the classifier genes alongside their top 5 relevant correlating genes is produced for each of the five different cases A-E leading to the obtention of classifier genes; see Table 6 . Table 6 − 1 . The classifier genes in the same row with their top 5 relevant correlating genes in case A classifier gene Gene_1 Gene_2 Gene_3 Gene_4 Gene_5 SEMA6A UNC13C TEX56P C10orf71-AS1 PCDH9 GTF2IP20 CDC26P1 FAM153A HMGN2 PCSK7 CAMKMT ENTREP3 CRADD-AS1 FAM98A CALM2P2 PCLO UCHL5 SLC26A3 VWC2L BACE1-AS CERS5 SUB1 HNRNPKP2 LINC03040 LRRC3B-AS1 GRAP ERAP1 C13orf46 P4HB APOBEC3C CYP3A43 OAS3 C6 TMEM86B INPP5F ABCB7 MIR5697 ACAD8 IRF1 LARS2 GIMAP8 FABP5 FAM178B LTA4H EHMT2 DHRS4 IMPACT HTRA3 CFAP107 SETD1B ISOC2 LIMD1 PDPR2P CRIM1-DT RPS2P32 LAP3 ZNF407 PAXIP1-DT ALDH9A1 SPIN4 NACA2 IL6R-AS1 NDUFB6 AGFG2 NUP188 C1orf127 HMGB1P37 NUDT12 RANBP6 LSR CYP3A5 TBC1D5 Table 6 − 2. The classifier genes in the same row with their top 5 relevant correlating genes in case B classifier gene Gene_1 Gene_2 Gene_3 Gene_4 Gene_5 C10orf71-AS1 TEX56P UNC13C GSCAR ITGA1 GTF2IP20 RANBP17 RNF135 MT-CO2 FAAH DMAP1 MORC2 CRADD-AS1 VN1R81P SPTBN1 GLOD4 TRIOBP SNHG9 CDC26P1 GSTZ1 RPRD2 DYNLT1 SEMA3E ERFE RTL9 INPP5F KIF6 GAPDHP70 COL5A1 MRPL33 CYP3A43 PET117 EFCAB14 FAM98A GOLGA6GP ZBTB44 MIR3682 TET1 UQCC2 WFS1 MORN2 NMB RPS2P32 IGDCC4 CCNB3 CDIP1 LIMD1-AS1 NAA35 CFAP107 PLCE1 RN7SL269P ENPP1 NBPF1 STEAP2 RNA5SP466 ANKLE1 AP1G2 BTG1-DT CNR1 EIF2B5-DT Table 6 − 3 . The classifier genes in the same row with their top 5 relevant correlating genes in case C Classifier gene Gene_1 Gene_2 Gene_3 Gene_4 Gene_5 C10orf71-AS1 STRADB SKAP2 OIP5-AS1 UNC13C ITGA1 CRADD-AS1 CLNS1A DRD4 ATG13 TM2D3 PSMD14 CDC26P1 SMIM10L1 USP15 LGR6 MRPL33 CCNY GSG1L LINC02937 SNORA33 ADCY10 RANBP10 CALML6 CYP3A43 NDUFAF1 TRNAU1AP COX18 ANKMY1 KIF9 LRRC3B-AS1 POLR2C ARVCF CTHRC1 EPB41L4B CENPV RPS2P32 KALRN CNTF PYROXD2 TNPO1 RBX1 CXCL3 CENPW SF3B6 ILRUN-AS1 MESDP1 SORT1 FAM238B NBDY FGGY-DT NDUFB10 OR7E53P TMEM79 CFAP107 HOMER3-AS1 PLEKHN1 C16orf86 HSPA8 PABPN1 Table 6 − 4 . The classifier genes in the same row with their top 5 relevant correlating genes in case D Classifier gene Gene_1 Gene_2 Gene_3 Gene_4 Gene_5 C10orf71-AS1 UNC13C SKAP2 TEX56P ITGA1 RPS27L CRADD-AS1 CBR1 ACTR5 PPIC TARS2 PPFIBP1 CYP3A43 KIAA2013 HMGXB4 KCNE2 CCNF KRT7 CDC26P1 CD27-AS1 ZNF580 MAP2K1 RPRD1A ZNF362 GSG1L TRIM52-AS1 KCNQ5 PHC1P1 THUMPD2 PPIH RANBP17 TSLP PDZD11 ALDH4A1 CD38 LINC03040 RTL9 FLCN MMP24OS NUDT11 NOTCH4 TPM2 RPS2P32 PCDHGA1 LIPT2-AS1 LSM12 RUBCNL CTDNEP1 CFAP107 MAN1A2 ZNF888 NLGN4Y PBX3 TAS2R20 Table 6 − 5 . The classifier genes in the same row with their top 5 relevant correlating genes in case E Classifier gene Gene_1 Gene_2 Gene_3 Gene_4 Gene_5 C10orf71-AS1 OIP5-AS1 UNC13C ITGA1 TEX56P STRADB CDC26P1 C1orf21 APBB3 PDP1 NOC2L SLC44A2 RTL9 MAP4K4 GUCY1A2 ATF7IP MCM7 CAST CRADD-AS1 SCOC BOD1 MAPKAPK2 SEPTIN5 LINC01311 LINC01121 ERG ETNPPL SPTBN4 ATP5IF1 GADD45G CYP3A43 CPNE1 AAMP RBM39 UBL3 NAB2 RANBP17 RENBP SLC48A1 SYNC VEZT GSPT1 RPS2P32 ATP5ME PIGC APOOL PMS2CL XK SHISA8 RAD23A DBT SYNE4 BAALC FDX1 RNA5SP466 CFAP20 LYSMD3 PARP2 RFPL3S ZKSCAN3 Procedures of kNN Sample Classification A validation dataset that is composed of the uncorrelated genes only for the validation sample is produced. Each validation sample is tested for frailty by calculating the Euclidean distance between the validation sample and an equal number of training frail and young samples that are randomly chosen. For each validation sample, a table of training samples and their Euclidean distances from the validation sample is produced. Also produced is the average of the distances of frail samples from the validation sample, and the average of the distances of young samples from the validation sample. In one classification scheme, the sample classifies as the phenotype of the nearest sample to it by Euclidean distance. In another classification scheme, the sample classifies by the phenotype of samples that are on average closer by Euclidean distance to it. Test Dataset The test dataset is composed of normalized and annotated counts of RNA-seq data obtained from PBMCs from middle-aged non-frail (mean age = 48) and frail (mean age = 47.8) African Americans and whites. The dataset was deposited in the Gene Expression Omnibus (GEO) database under the title “Frailty in middle age is associated with race-specific changes to the transcriptome”, made public in August 21, 2019, and accessible from the Gene Expression Omnibus (GEO) with accession number GSE129534. The total number of samples in the dataset is 16, with each sample coming from a separate individual. The individuals are phenotypically categorized as frail or non-frail. Declarations Author Contribution The entire work featuring in the manuscript has been authored by a single individual, Marshall Tabetah, initialized as MT.Conceptualization: M.T.; Methodology: M.T.; Data Processing & Analysis: M.T.; Investigation: M.T.; Original Draft: M.T.; Review & Editing: M.T.; Figures and Visualization: M.T.; Supervision: M.T.; Funding: The authors received no funding for this work Acknowledgement Marshall Tabetah acknowledges Genelab AWG Frailty Group for useful discussions that motivated a search for frailty classifier genes. Data Availability The dataset that is used to construct the set of classifier genes is raw counts of RNA-seq data obtained from muscle biopsies of vastus lateralis, deposited in the Gene Expression Omnibus (GEO) database under the title “Muscle function in fit and (pre-)frail males and females”, made public in Feb 04, 2023, and accessible from the Gene Expression Omnibus (GEO) with accession number [GSE144304](https:/www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE144304) . The RNA-seq counts were obtained with the Illumina NextSeq 500 (Homo sapiens) platform from samples that were submitted in Jan 27, 2020. The total number of samples in the dataset is 80, with each sample coming from a separate individual. The individuals are phenotypically categorized as young, fit, or frail. Old subjects ( > 75 years) were screened for frailty using the Fried criteria. (e.g. weight loss, exhaustion, low physical activity, slowness and weakness) []. References D.B. Dongare, S.S. Nishad, S.Y. Mastoli, S.A. Saraf, N. Srivastava, A. Dey, High throughput sequencing: a breakthrough in molecular diagnosis for precision medicine, Functional & Integrative Genomics (2025) 25:22 https://doi.org/10.1007/s10142-025-01529-w J.M. Churko, G.L. Mantalas, M.P. Snyder, and J.C. Wu, Overview of High Throughput Sequencing Technologies to Elucidate Molecular Pathways in Cardiovascular Diseases, Circ Res. 2013 June 7; 112(12): doi: 10.1161/CIRCRESAHA.113.300939 J-Y Lee, The Principles and Applications of High-Throughput Sequencing Technologies, Dev Reprod 2023;27(1):9–24 https://doi.org/10.12717/DR.2023.27.1.9 Amer F.A.H. ALNUAIMI and Tasmin H.K. ALBALDAWI, An overview of machine learning classification techniques, BIO Web of Conferences 97 , 00133 (2024) ISCKU 2024 https://doi.org/10.1051/bioconf/20249700133 Z. Yaping, Z. Changyin, Gene Feature Selection Method Based on ReliefF and Pearson Correlation, 2021 3rd International Conference on Applied Machine Learning (ICAML). DOl 1O.1109/1CAML54311.2021.00011 X. Cao, L. Ding and T.B. Mersha, Development and validation of an RNA‑seq‑based transcriptomic risk score for asthma, Scientific Reports (2022) 12:8643 https://doi.org/10.1038/s41598-022-12199-0 Bostanci, E.; Kocak, E.; Unal, M.; Guzel, M.S.; Acici, K.; Asuroglu, T. Machine Learning Analysis of RNA-seq Data for Diagnostic and Prognostic Prediction of Colon Cancer. Sensors 2023 , 23, 3080. https://doi.org/10.3390/s23063080 Z. Yaping, Z. Changyin, Gene Feature Selection Method Based on ReliefF and Pearson Correlation, 2021 3rd International Conference on Applied Machine Learning (ICAML). DOI1O.1109/1CAML54311.2021.00011 J. Perez-Ortega, N.N. Almanza-Ortega, A. Vega-Villalobos, R. Pazos-Rangel, C. Zavala-Diaz, and A. Martinez-Rebollar, The K-Means Algorithm Evolution, DOI : http://dx.doi.org/10.5772/intechopen.85447 R.J. Urbanowicz, M. Meeker, W. La Cava, R.S. Olson, J.H. Moore, Relief-based feature selection: Introduction and review, Journal of Biomedical Informatics 85 (2018) 189–203 https://doi.org/10.1016/j.jbi.2018.07.014 M. I. Love, W. Huber, S. Anders: Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology 2014, 15 :550. http://dx.doi.org/10.1186/s13059-014-0550-8 G.J. Szekely, M.L. Rizzo, and N.K. Bakirov, Measuring and Testing Dependence by Correlation of Distances, The Annals of Statistics 2007, Vol. 35, No. 6, 2769–2794. DOI: 10.1214/009053607000000505 J. Sun, J.D. Herazo-Maya, X. Huang, N. Kaminski, H. Zhao, Distance correlation based gene set analysis in longitudinal studies, Statistical Applications in Genetics and Molecular Biology. 2018; 20170053. Hou et al. Distance correlation application to gene co-expression network analysis, BMC Bioinformatics (2022) 23:81 https://doi.org/10.1186/s12859-022-04609-x R.A. Tumasian III, A. Harish, G. Kundu, J-H Yang, C. Ubaida-Mohen, M. Gonzalez-Freire, M. Kaileh, L.M. Zukley, C.W. Chia, A. Lyashkow, W.H. Wood III, Y. Piao, C. Coletta, J. Ding, M. Gorospe, R. Sen, S. De & L. Ferruci, Skeletal muscle transcriptome in healthy aging, NATURE COMMUNICATIONS (2021) 12:2014 https://doi.org/10.1038/s41467-021-22168-2 Fried, L.P., Tangen, C.M., Walston, J., et al. (2001) Cardiovascular Health Study Collaborative Research Group. Frailty in Older Adults: Evidence for a Phenotype. The Journals of Gerontology Series A: Biological Sciences and Medical Sciences, 56, M146-M156. http://dx.doi.org/10.1093/gerona/56.3.M146 M.I. Love, S. Anders, W. Huber: Differential analysis of count data – the DESeq2 package (2016) Additional Declarations No competing interests reported. Supplementary Files overlapE.tsv overlapA.tsv overlapB.tsv overlapD.tsv overlapC.tsv Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7475052","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":534278351,"identity":"7a587694-cdb4-4987-b7a7-fd1e8b592562","order_by":0,"name":"Marshall Tabetah","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1ElEQVRIiWNgGAWjYJACA4YKBh4GBsYGhgQQ9wBBDcwMBQfOkKrlw8E2ZAFCWnTb+w9u/jjvsIx5/+K2Bw9qGOT4biTg12J25jCzwcFth3lkbjxsN0g4xmAsSVDLjWQ2oJY0HgmJg20SCWwMiRsIarn/mP3HwTkwLf8Y6glrucHMYHCwwYZHgr+xTSKxjSHBgLBfkg0MzhwDapFgBGrpkzCceeYBAS3HDz4wqKiRsJfgP/5M8sc3G3m+4wRsQQAJsEoJYpWDAP8BUlSPglEwCkbBSAIAK9VImZ7IYMoAAAAASUVORK5CYII=","orcid":"","institution":"University of Buea","correspondingAuthor":true,"prefix":"","firstName":"Marshall","middleName":"","lastName":"Tabetah","suffix":""}],"badges":[],"createdAt":"2025-08-28 00:08:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7475052/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7475052/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":94445931,"identity":"25280a78-f8db-4069-b4f7-272824239b97","added_by":"auto","created_at":"2025-10-27 14:33:46","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2495174,"visible":true,"origin":"","legend":"","description":"","filename":"SubmissionNatureAgingFrailtyClassifierGenesPaper.docx","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/551be850779c3aa8b8de79d1.docx"},{"id":94446072,"identity":"bb42c164-b1d6-45e0-849c-3044ebdf2913","added_by":"auto","created_at":"2025-10-27 14:34:00","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4142,"visible":true,"origin":"","legend":"","description":"","filename":"94324a1ce4354c4b9fe15f6e5755f0cd.json","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/b28b0749872034f7ad26230d.json"},{"id":94446073,"identity":"75ab1116-f4af-4d64-90e4-579078d364e4","added_by":"auto","created_at":"2025-10-27 14:34:00","extension":"tsv","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":88487,"visible":true,"origin":"","legend":"","description":"","filename":"overlapA.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/902be46ed03b179f20b337f5.tsv"},{"id":94445951,"identity":"3e2262e1-483f-4c46-93e1-85d512d6cfd2","added_by":"auto","created_at":"2025-10-27 14:33:50","extension":"tsv","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":87640,"visible":true,"origin":"","legend":"","description":"","filename":"overlapB.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/6d19c689c3ab0a07ebf2b724.tsv"},{"id":94445346,"identity":"6a9e4c6f-23ca-4c88-8ac3-c37e2b8cdde7","added_by":"auto","created_at":"2025-10-27 14:33:00","extension":"tsv","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":86951,"visible":true,"origin":"","legend":"","description":"","filename":"overlapC.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/1c40b198e91bb8f21ae41099.tsv"},{"id":94446080,"identity":"b622b167-4a3c-4935-814e-b202930206eb","added_by":"auto","created_at":"2025-10-27 14:34:01","extension":"tsv","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":82605,"visible":true,"origin":"","legend":"","description":"","filename":"overlapD.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/364adc96276e68447cbc9753.tsv"},{"id":94445719,"identity":"087610d6-acc4-47e2-ae78-f4830fb67f86","added_by":"auto","created_at":"2025-10-27 14:33:29","extension":"tsv","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":85653,"visible":true,"origin":"","legend":"","description":"","filename":"overlapE.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/91604a5ce315d4b3030f3392.tsv"},{"id":94445869,"identity":"98444e71-344b-44d6-aba2-088de96e4eaa","added_by":"auto","created_at":"2025-10-27 14:33:41","extension":"xml","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":176266,"visible":true,"origin":"","legend":"","description":"","filename":"94324a1ce4354c4b9fe15f6e5755f0cd1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/4dd369fe98eeaba578aaec60.xml"},{"id":94445885,"identity":"cbefd7c2-5f20-43b7-ae64-b648559e19b7","added_by":"auto","created_at":"2025-10-27 14:33:42","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":186408,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/25a8853ab23ac7fbc2b8b037.jpeg"},{"id":94445945,"identity":"3dfa1f94-ca78-449b-a692-4929756d6235","added_by":"auto","created_at":"2025-10-27 14:33:49","extension":"emf","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2442744,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.emf","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/b31543cbdfeaaa0e96977cd8.emf"},{"id":94490016,"identity":"5b9b0905-3e00-4cc5-b073-39a19b4cc752","added_by":"auto","created_at":"2025-10-27 17:07:00","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":105380,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/da939ec1f9666f9ecc27d159.jpeg"},{"id":94445403,"identity":"a03eabc8-16b3-4e70-a148-ea95f2117270","added_by":"auto","created_at":"2025-10-27 14:33:08","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":107220,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage12.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/066409b156f4858161211be7.jpeg"},{"id":94445918,"identity":"5bf2722b-473a-4aaf-9acc-d8bd22d614a6","added_by":"auto","created_at":"2025-10-27 14:33:44","extension":"emf","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1478736,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage13.emf","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/b1ecbbbe0cb8bcb023cda3da.emf"},{"id":94445617,"identity":"3e3ecca0-9677-4318-b991-1d8c336fd4f9","added_by":"auto","created_at":"2025-10-27 14:33:21","extension":"emf","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1479312,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage14.emf","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/9cab018d362a235888cf1ae3.emf"},{"id":94446005,"identity":"cd260b8a-1e44-4837-b176-89ce485335c4","added_by":"auto","created_at":"2025-10-27 14:33:54","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":91428,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/9a4c8ab627ff49b58a809a99.png"},{"id":94446090,"identity":"302b8860-55e2-4e83-a172-e6ac3de33d54","added_by":"auto","created_at":"2025-10-27 14:34:05","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":46689,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/9734dcca99199d14c0a32a99.png"},{"id":94445936,"identity":"03e9c13c-b701-4a16-8bb2-ffcfd28d30fa","added_by":"auto","created_at":"2025-10-27 14:33:46","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13496,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage17.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/b2e7fc644e4b57c743218152.png"},{"id":94445205,"identity":"cb38bf2d-9878-430a-b3b7-93c2f8b50bab","added_by":"auto","created_at":"2025-10-27 14:32:55","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15860,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage18.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/9bc7dfe9e3252f2c74566de0.png"},{"id":94445608,"identity":"afac7592-b819-4d09-9372-13816877b8e9","added_by":"auto","created_at":"2025-10-27 14:33:20","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":46689,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/7df53c2f85b9ef78c431ec91.png"},{"id":94445717,"identity":"64d3aaf6-712a-411e-acf3-9eb989bdf941","added_by":"auto","created_at":"2025-10-27 14:33:28","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":34495,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/7cd94921b81b20cab3d1323c.png"},{"id":94445610,"identity":"e9ede418-a398-47de-a376-604430f5f271","added_by":"auto","created_at":"2025-10-27 14:33:21","extension":"jpeg","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":925392,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/7b7bfbd8aa6b7cc237bd0b52.jpeg"},{"id":94445399,"identity":"54927876-c78b-473d-b7ad-94109b6f38db","added_by":"auto","created_at":"2025-10-27 14:33:08","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":34244,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/4b4024ba0b5871390407676f.png"},{"id":94445590,"identity":"5540fc71-6186-4863-ba0e-4bacc3307f5e","added_by":"auto","created_at":"2025-10-27 14:33:16","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33261,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/4391a82a47b7b59c7f1470d9.png"},{"id":94445953,"identity":"52e668eb-8f54-41a0-9bd2-fb9a65a08a48","added_by":"auto","created_at":"2025-10-27 14:33:50","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23857,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/dea2032d2e9562a6f15fb549.png"},{"id":94445933,"identity":"2199be12-e9b9-44cb-b127-1eebbac122a0","added_by":"auto","created_at":"2025-10-27 14:33:46","extension":"jpeg","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":545424,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/6fb7270f83aeda161128c42c.jpeg"},{"id":94445358,"identity":"bc2b65d0-ff0b-42af-a2a1-d8135d73f7be","added_by":"auto","created_at":"2025-10-27 14:33:06","extension":"jpeg","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":107594,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/6e51af99732040d18a258fcd.jpeg"},{"id":94445975,"identity":"e9664942-a1bf-4dca-94c8-6d6fb5086cc4","added_by":"auto","created_at":"2025-10-27 14:33:51","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":28165,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/a9441df10df9555e98f43497.png"},{"id":94446078,"identity":"a0000074-e526-49b7-9ee7-6d3e12de2608","added_by":"auto","created_at":"2025-10-27 14:34:01","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":20433,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/59a3550edf69d7402e077948.png"},{"id":94445651,"identity":"887eef3f-1edc-4799-b71c-44a3d68de0bf","added_by":"auto","created_at":"2025-10-27 14:33:23","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18517,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/51c102d4bcec797a130414a9.png"},{"id":94445591,"identity":"e790b3e5-409b-4453-97c5-cd75f6a7306c","added_by":"auto","created_at":"2025-10-27 14:33:16","extension":"png","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18868,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/ae987ac0c3d192518ef27f63.png"},{"id":94445877,"identity":"d975e7d1-7ca6-4077-bba4-96812005da45","added_by":"auto","created_at":"2025-10-27 14:33:42","extension":"png","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16700,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/d73447ed192a5cb0e2974391.png"},{"id":94446089,"identity":"7fbd3031-692c-4d70-a241-bbfcc22a095b","added_by":"auto","created_at":"2025-10-27 14:34:05","extension":"png","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16801,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/d0944b84f20946088116ae2d.png"},{"id":94446081,"identity":"95e7ef24-e547-47ca-ad97-70cf8116f2d4","added_by":"auto","created_at":"2025-10-27 14:34:01","extension":"png","order_by":32,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31870,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/f7e5c8fa04f2855f9051534d.png"},{"id":94445401,"identity":"a01f5668-4e65-4b5b-b4e5-8089641d6459","added_by":"auto","created_at":"2025-10-27 14:33:08","extension":"png","order_by":33,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47566,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/ea5ac0e0c22035fa4b3e5394.png"},{"id":94445949,"identity":"9455005e-90ab-4361-a69c-ffd9bf15587b","added_by":"auto","created_at":"2025-10-27 14:33:50","extension":"png","order_by":34,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14850,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage17.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/fe1dd115c9a6c9d32af8e157.png"},{"id":94445111,"identity":"96e3738a-47d2-4781-8573-3d731e613d2b","added_by":"auto","created_at":"2025-10-27 14:32:47","extension":"png","order_by":35,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17029,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage18.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/450127c852baf834f8881d3c.png"},{"id":94445663,"identity":"51366531-11df-4fbd-bc82-e45e19ea14f8","added_by":"auto","created_at":"2025-10-27 14:33:25","extension":"png","order_by":36,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47566,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/8fd95af5eb28627713b89f82.png"},{"id":94446092,"identity":"e43263b0-c605-4db1-8dee-9271b7a3bffb","added_by":"auto","created_at":"2025-10-27 14:34:06","extension":"png","order_by":37,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33582,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/a344e619e18dde494fd4006a.png"},{"id":94444987,"identity":"ae7a843d-1990-4c65-a8bb-9e9a122a5aed","added_by":"auto","created_at":"2025-10-27 14:32:36","extension":"png","order_by":38,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7777,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/0055a34cd99686a33dc60c83.png"},{"id":94445922,"identity":"433482d3-16ad-4b2c-9119-a11a2f1c374c","added_by":"auto","created_at":"2025-10-27 14:33:44","extension":"png","order_by":39,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33135,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/b58c7a0add30fac66332a4cd.png"},{"id":94489344,"identity":"8c0425b2-be83-45ae-840f-0e7fcc762bcb","added_by":"auto","created_at":"2025-10-27 17:04:15","extension":"png","order_by":40,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32363,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/aa1070ae82fc7f76db0a497a.png"},{"id":94445720,"identity":"b8359bea-2ee3-49f4-aec0-797b2d2dc79c","added_by":"auto","created_at":"2025-10-27 14:33:29","extension":"png","order_by":41,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25011,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/718d5c48d36e322985ac51c7.png"},{"id":94446004,"identity":"ff65f635-7935-4b04-b886-2dfb50a40d21","added_by":"auto","created_at":"2025-10-27 14:33:54","extension":"png","order_by":42,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4239,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/8555cf2e5c181a6b740abd82.png"},{"id":94445208,"identity":"8f5a1b90-d981-4c20-8029-76fe84d75fa1","added_by":"auto","created_at":"2025-10-27 14:32:55","extension":"png","order_by":43,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18923,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/019790754aaf816d1404be82.png"},{"id":94446071,"identity":"5783127b-6deb-43ef-a111-5fdcee2a90a2","added_by":"auto","created_at":"2025-10-27 14:33:59","extension":"xml","order_by":44,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":174824,"visible":true,"origin":"","legend":"","description":"","filename":"94324a1ce4354c4b9fe15f6e5755f0cd1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/25dd29abf7b20e8ac078e967.xml"},{"id":94446037,"identity":"70dc8c6f-d50e-4ae2-906a-3a07176d3cbd","added_by":"auto","created_at":"2025-10-27 14:33:58","extension":"html","order_by":45,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":193602,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/930800c5aed70098f4c3c296.html"},{"id":94445348,"identity":"7286cf27-0b75-4a68-ba2b-5ca5ed94fee9","added_by":"auto","created_at":"2025-10-27 14:33:00","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":135375,"visible":true,"origin":"","legend":"\u003cp\u003eThe stages of the data mining pipeline leading to obtention of classifier genes and sample classification.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/ba3f4f02ed33a5a95a9ff557.png"},{"id":94445743,"identity":"60e9a211-a242-467a-a563-bf7f92d4a09f","added_by":"auto","created_at":"2025-10-27 14:33:35","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":143952,"visible":true,"origin":"","legend":"\u003cp\u003eMDS plot of 48 young and frail samples, after extreme outliers \u003cem\u003es525_F_Yo\u003c/em\u003e and \u003cem\u003es326_M_Fr\u003c/em\u003e are removed. Some samples depicted as outliers can be clearly seen to be in the midst of the contrasting phenotype, as with\u003cem\u003e s323_M_Fr\u003c/em\u003e, \u003cem\u003es514_F_Yo\u003c/em\u003e; and the samples \u003cem\u003es508_M_Yo\u003c/em\u003e, and \u003cem\u003es530_F_Yo \u003c/em\u003ethat seem surrounded by samples of both phenotypes are closer to frail samples than young samples.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/a7a112ac590d1973dc98e4b9.png"},{"id":94445893,"identity":"60cc52ae-7090-407f-8963-305d6d69381d","added_by":"auto","created_at":"2025-10-27 14:33:42","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":187459,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 3-1\u003c/strong\u003e. Diagram indicating the cluster of each sample in the set of 44 samples that consists of 22 young samples and 22 frail samples, based on the entire set of 20638 genes. All young samples are in cluster 1, and all frail samples, apart from \u003cem\u003es302_M_Fr\u003c/em\u003e, \u003cem\u003es303_M_Fr\u003c/em\u003e, \u003cem\u003es306_M_Fr\u003c/em\u003e, and \u003cem\u003es318_F_Fr\u003c/em\u003e, are in cluster 2. The illustrated representation is maintained over 12 independent runs of the k-means clustering algorithm.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/ae8afed56b9c1a881005a327.png"},{"id":94445609,"identity":"c8fea058-93d9-4f24-baa1-56ff1bf28692","added_by":"auto","created_at":"2025-10-27 14:33:20","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":179135,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 3-2\u003c/strong\u003e. Diagram indicating the cluster of each sample in the set of 44 samples that consists of 22 young samples and 22 frail samples in a dataset of 37 classifier frailty genes only. All frail samples, apart from \u003cem\u003es306_M_Fr\u003c/em\u003e, are in one cluster indexed as 2; and all young samples, apart from samples \u003cem\u003es512_M_Yo\u003c/em\u003e, \u003cem\u003es527_M_Yo\u003c/em\u003eand \u003cem\u003es528_M_Yo\u003c/em\u003e, are in one cluster indexed as 1. The illustrated representation is maintained over 12 independent runs of the k-means clustering algorithm.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/9c1262cc9e7ec75393d5db53.png"},{"id":94445506,"identity":"fb375aee-3c60-4d7f-8fb3-5ba9daec35ab","added_by":"auto","created_at":"2025-10-27 14:33:15","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":151542,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 4-1\u003c/strong\u003e. Diagram indicating the cluster of each sample in the set of 40 samples that consists of 22 young samples and 18 frail samples, based on the entire set of 24238 genes. All young samples are in cluster 1, and all frail samples are in cluster 2. The illustrated representation is maintained over 12 independent runs of the k-means clustering algorithm.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/e5ffdaea4af18d487a7620f5.png"},{"id":94446076,"identity":"59ad5d7d-888c-47e3-8fc7-720d53d64ca3","added_by":"auto","created_at":"2025-10-27 14:34:01","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":189706,"visible":true,"origin":"","legend":"\u003cp\u003eFigure 4-2. Diagram indicating the cluster of each sample in the set of 40 samples that consists of 22 young samples and 18 frail samples in a dataset of classifier frailty genes only, obtained at δW=0.01, R_th=0.4. All frail samples are in one cluster indexed as 1, and all young samples, apart from samples s516_F_Yo and s522_F_Yo, are in one cluster indexed as 2. The illustrated representation is maintained over 12 independent runs of the k-means clustering algorithm.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/2b72067c1436ac9b6076af71.png"},{"id":94445716,"identity":"b8bff7f6-4ca4-4774-8d93-3a426f4c53db","added_by":"auto","created_at":"2025-10-27 14:33:28","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":60394,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 5-1\u003c/strong\u003e. MDS plot of 8 young and 8 frail samples, based on the entire gene-set of the test dataset.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/27a7d0ef75afeb6b9ee44896.png"},{"id":94445359,"identity":"c8305747-c737-416b-9bda-2794671e8a22","added_by":"auto","created_at":"2025-10-27 14:33:07","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":103432,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 5-2\u003c/strong\u003e. Diagram indicating the cluster of each sample in the test dataset of 16 samples that is based on the original gene-set. The illustrated representation is maintained over 4 independent runs of the k-means clustering algorithm.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/0a0768ef7b319ab21198babc.png"},{"id":94445938,"identity":"c2113a31-ea5a-4bba-9087-93a8390a5223","added_by":"auto","created_at":"2025-10-27 14:33:47","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":162339,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 6\u003c/strong\u003e. Identification of PBMC middle-aged frail and non-frail samples in 4 clusters generated by k-means. Frail samples are identified in one color, and non-frail samples are identified in another color. Panel \u003cstrong\u003ea\u003c/strong\u003e) is based on the entire gene-set of the PBMC middle-aged samples; \u003cstrong\u003eb\u003c/strong\u003e) is based on the 10 classifier genes of the case \u003cstrong\u003eC\u003c/strong\u003e scenario of classifier genes identification in the training skeletal muscle dataset.\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/b61ebd260b7bb5201b538855.png"},{"id":94446088,"identity":"d658ab5d-d60d-4f11-a033-285658f35ebc","added_by":"auto","created_at":"2025-10-27 14:34:04","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":223091,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 7\u003c/strong\u003e. Display of PBMC middle-aged frail and non-frail samples in 4 clusters generated by k-means, based on classifier genes alone. Frail samples are identified in one color, and non-frail samples are identified in another color. Panels a) – d) represent clusters obtained with classifier gene sets in cases \u003cstrong\u003eB\u003c/strong\u003e, \u003cstrong\u003eA\u003c/strong\u003e, \u003cstrong\u003eD\u003c/strong\u003e, and \u003cstrong\u003eE\u003c/strong\u003e, respectively.\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/9e5f59fef2f1cce5cf7a451b.png"},{"id":94446038,"identity":"af455fb2-0617-4b96-b6d6-51feb292b64a","added_by":"auto","created_at":"2025-10-27 14:33:59","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":135341,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 7. \u003c/strong\u003eMDS plot of all young and frail samples. Here s525_F_Yo and s326_M_Fr seem outstandingly distant from the rest.\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/1eca81f5cd25550d29f09eef.png"},{"id":94445722,"identity":"182c1654-1437-4b6d-b71d-5bcea5fe37f1","added_by":"auto","created_at":"2025-10-27 14:33:29","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":142713,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 8\u003c/strong\u003e. MDS plot of 48 young and frail samples, with s525_F_Yo and s326_M_Fr removed.\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/f1cc8f0dc4ce5968cdfd3118.png"},{"id":94445204,"identity":"7ca6c599-3bd6-4ea7-9d40-478c78a37aaf","added_by":"auto","created_at":"2025-10-27 14:32:54","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":33361,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 9. \u003c/strong\u003eA graph of the density of samples in the neighborhood of each sample against the sample index. Indices 19 and 40, corresponding to samples s326_M_Fr and s_525_F_Yo, respectively, show only 1 sample in their neighborhood.\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/316ae63b36dea48cd0f7a3fc.png"},{"id":94489635,"identity":"457663c9-23d4-49ec-852b-c525bb0b3e1c","added_by":"auto","created_at":"2025-10-27 17:05:27","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":37042,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFigure 10. \u003c/strong\u003eA graph of the density of samples in the neighborhood of each sample against the sample index for 48 samples after retrieval of samples s326_M_Fr and s_525_F_Yo. Index 35, corresponding to sample s_514_F_Yo shows only 6 samples in its neighborhood.\u003c/p\u003e","description":"","filename":"14.png","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/7daa33e490d8ea02751a96d9.png"},{"id":94505710,"identity":"9f86b57f-b843-4d34-8a47-c640d272a8a8","added_by":"auto","created_at":"2025-10-28 16:21:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3612007,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/03c5e58c-17b8-4b1c-bbe4-4b41c645c825.pdf"},{"id":94445948,"identity":"19876f39-b7ad-4d23-a7d5-ef497df4123c","added_by":"auto","created_at":"2025-10-27 14:33:49","extension":"tsv","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":85653,"visible":true,"origin":"","legend":"","description":"","filename":"overlapE.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/d3068d4a0ee0572fdb326bec.tsv"},{"id":94445109,"identity":"4c605126-e863-4300-94de-2ed38001c807","added_by":"auto","created_at":"2025-10-27 14:32:46","extension":"tsv","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":88487,"visible":true,"origin":"","legend":"","description":"","filename":"overlapA.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/b22dce33ae36d008e6456dcd.tsv"},{"id":94445203,"identity":"5832e4a1-5ca4-49e2-b671-46f2b1269c9c","added_by":"auto","created_at":"2025-10-27 14:32:54","extension":"tsv","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":87640,"visible":true,"origin":"","legend":"","description":"","filename":"overlapB.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/c98678006f6c4278c1b54f90.tsv"},{"id":94445932,"identity":"02c1e6b2-c818-494a-97bb-ff167b859282","added_by":"auto","created_at":"2025-10-27 14:33:46","extension":"tsv","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":82605,"visible":true,"origin":"","legend":"","description":"","filename":"overlapD.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/67b05e7b0ce9733848a66a3b.tsv"},{"id":94445586,"identity":"936a6cdb-9a0e-44c4-9ae0-81bd194d8400","added_by":"auto","created_at":"2025-10-27 14:33:16","extension":"tsv","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":86951,"visible":true,"origin":"","legend":"","description":"","filename":"overlapC.tsv","url":"https://assets-eu.researchsquare.com/files/rs-7475052/v1/c157d1580be51e0a2782b105.tsv"}],"financialInterests":"No competing interests reported.","formattedTitle":"RNA-Seq Data Analyses and Machine Learning Predict Frailty Classifier Genes","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eRNA sequencing is a cost-effective high throughput sequencing (HTS) technology that sequences the whole genome with remarkable speed, throughput and accuracy [\u003ca class=\"FNLink\" href=\"#Fn1\" id=\"#FNLinkFn1\"\u003e\u003c/a\u003e, \u003ca class=\"FNLink\" href=\"#Fn2\" id=\"#FNLinkFn2\"\u003e\u003c/a\u003e, \u003ca class=\"FNLink\" href=\"#Fn3\" id=\"#FNLinkFn3\"\u003e\u003c/a\u003e]. RNA-seq is used to quantify gene expression levels, and this has enabled its importance and application in the diagnosis and early detection of disease or pathological conditions. RNA-sequencing samples are characteristic of a high number of genes, typically in thousands or tens of thousands. However, only a subset of the genes may be related to a specific condition, making the remaining genes irrelevant. Among the relevant genes are genes that exercise some correlation across the samples in relation to the specific disease or condition. Genes that weakly correlate with one another can be used to accurately classify the condition. The remainder of genes that do not survive among the weakly correlated are thus deemed redundant. The selection method thus involves filtering out genes that are irrelevant, and then removing any redundant genes, in order to end up with a small set of genes that can adequately classify the pathological condition of interest.\u003c/p\u003e\u003cp\u003eObtaining a small set of genes from the sequenced gene-set is called dimension reduction in the literature [\u003ca class=\"FNLink\" href=\"#Fn4\" id=\"#FNLinkFn4\"\u003e\u003c/a\u003e, \u003ca class=\"FNLink\" href=\"#Fn5\" id=\"#FNLinkFn5\"\u003e\u003c/a\u003e]. In this perspective, genes are inherently seen as independent features, implying that the expression of a gene is in a dimension that is unique from the remaining genes. Reducing the genes to a small subset is thus resultant in a dramatic reduction in the number of dimensions initially available.\u003c/p\u003e\u003cp\u003eModels that identify biomarker genes for predicting the prevalence of disease have been developed, as a few examples can be named: a RNA-seq transcriptomic risk score for predicting asthma, based on 73 differentially expressed genes (DEG) using Logistic least absolute shrinkage and selection operator regression analysis (Lasso) [\u003ca class=\"FNLink\" href=\"#Fn6\" id=\"#FNLinkFn6\"\u003e\u003c/a\u003e]; the use of canonical machine learning (ML) classifiers such as k-Nearest Neighbor (kNN) and random forest (RF), and 1-D convolutional neural network (CNN), to predict colon cancer [\u003ca class=\"FNLink\" href=\"#Fn7\" id=\"#FNLinkFn7\"\u003e\u003c/a\u003e]; the use of a gene feature selection approach for dimensional reduction to a small number of classifier genes, and subsequent application of ML classifiers kNN and support vector machine (SVM) to classify sample data [\u003ca class=\"FNLink\" href=\"#Fn8\" id=\"#FNLinkFn8\"\u003e\u003c/a\u003e]. Feature selection is either an integral part of, or coupled with, the predictive model in some approaches, as in Refs [6, 7]. In other approaches, the feature selection is independent of the classification procedure that determines the phenotype, as in Ref. [8]. Another popular category of ML classifiers are clustering methods, which are unsupervised learning techniques unlike the aforementioned techniques which are categorized as supervised learning techniques. A clustering method that is implemented for classification in this work is k-means \u0026ndash; one of the most used clustering algorithms [\u003ca class=\"FNLink\" href=\"#Fn9\" id=\"#FNLinkFn9\"\u003e\u003c/a\u003e].\u003c/p\u003e\u003cp\u003eIn this work, a feature reduction algorithm called ReliefF [5, \u003ca class=\"FNLink\" href=\"#Fn10\" id=\"#FNLinkFn10\"\u003e\u003c/a\u003e] is used to determine genes that are relevant to frailty in a dataset of frail and non-frail samples. Redundant genes among the relevant genes are removed using distance correlation. The genes that have survived irrelevance elimination and redundancy elimination are termed classifier genes. The subset of samples that are used to obtain the classifier genes are \u003cem\u003etraining samples\u003c/em\u003e, while the remaining samples of the dataset are \u003cem\u003evalidation samples\u003c/em\u003e. kNN is used as a classifier method to test the validation samples based on the classifier genes. An alternative method (unsupervised learning approach) used for testing of samples is k-means clustering. \u003cem\u003ek-means\u003c/em\u003e is used for verifying whether samples of contrasting phenotypes cluster in separate groups. k-means is indispensable for testing samples in test datasets that are not the same as the training and validation dataset. kNN or an alternative supervised learning approach cannot be used for testing across datasets without account of batch effects.\u003c/p\u003e\u003cp\u003eThe data mining pipeline that involves the different stages of data processing, data analyses, and machine learning for gene selection and sample classification is displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eThere are a number of interesting findings from the investigations. The outcome of classification of validation samples based on classifier genes in a supervised learning approach and in an unsupervised learning approach can show up to be the same. With only\u0026thinsp;~\u0026thinsp;10 classifier genes, the classification of samples closely reproduces the outcome of classification with the entire gene-set made of tens of thousands of genes: classification with a size reduction factor of more than 1000 is thus achieved. Classification across a test dataset of frail and non-frail samples by k-means shows that there is no clear division between frail and non-frail samples. The picture provided by the k-means clustering is similar to the distribution of samples in a multi-dimensional scaling plot that shows frail and non-frail samples interspersed with one another. When a 4-centroid k-means approach is adopted, most of the samples in any cluster are of the same phenotype, making the belonging of the test samples more discernible.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"RESULTS AND DISCUSSIONS","content":"\u003cp\u003eA dataset of human frail and non-frail samples leveraged from the Gene Expression Omnibus (GEO) database is used for determining classifier genes of frailty, by using a sub-set of the data to train the gene selection algorithm, and the remainder of samples to validate the classifier genes. The dataset is a table of raw counts data made of 80 samples that is read into RStudio 4.3. Among the 80 samples there are male and female samples. Only groups of young males, frail males, young females, and frail females are of interest here, as the author wishes to properly distinguish between non-frail (represented by young samples) and frail samples. A subset of all young and frail samples is used as training samples for performing the selection of genes that are uniquely relevant to frailty. The select genes are later used to test whether a newly introduced sample from the remaining subset of samples is frail or not. The remaining subset of samples that are used to test frailty are called validation samples. Testing for a condition within the same dataset of samples is called validation.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eData Processing: Identification of Outlier Samples, Outlier Gene Counts, and Normalization of Gene Counts\u003c/h2\u003e\u003cp\u003eThe training dataset is composed as a set of two phenotypically distinct groups of young and frail samples. There is a total of 24 frail samples and a total of 26 young samples, making a total of 50 samples. After removing any outlying samples, the surviving samples are split into a training set and a validation set.\u003c/p\u003e\u003cp\u003eThe script DESeq2.R, as described in Ref. [\u003ca class=\"FNLink\" href=\"#Fn11\" id=\"#FNLinkFn11\"\u003e\u003c/a\u003e], is used as a reference to run code that detects any outlier counts and performs normalization of the gene counts in the dataset of 50 samples. After genes with a sum of counts less than 10 across all samples are removed, the number of surviving genes is 24993. Fifty-one (51) gene counts are identified as outliers, and replaced by modified counts.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eIdentification of Outlier Samples based on visualization of MDS plots\u003c/h3\u003e\n\u003cp\u003eThe essence of detection and removal of outlying samples is to remain with two groups of samples that are not only phenotypically distinct, but also close to being genotypically distinct, to enable the selection of genes with expressions that show a proper contrast between the two groups. As a prior attempt to visibly identify outlying samples, multidimensional scaling (MDS) is applied to the dataset of normalized counts to see the similarity of samples by their leading log Fold Change (logFC) in a MDS plot. The distances between samples correlate with the dissimilarity between samples in a MDS plot. Samples that seem to be visibly closer to samples of the phenotypically contrasting condition than to samples of the same phenotypic condition are potential outliers. By visualization of MDS plots, the frail samples deemed to be outliers and removed are \u003cem\u003es326_M_Fr\u003c/em\u003e, and \u003cem\u003es323_M_Fr\u003c/em\u003e; and for the young samples \u0026ndash; \u003cem\u003es525_F_Yo\u003c/em\u003e, \u003cem\u003es514_F_Yo\u003c/em\u003e, \u003cem\u003es508_M_Yo\u003c/em\u003e, and \u003cem\u003es530_F_Yo\u003c/em\u003e. The MDS plot of samples after removal of extreme outliers \u003cem\u003es525_F_Yo\u003c/em\u003e and \u003cem\u003es326_M_Fr\u003c/em\u003e is displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Most of the frail samples are located in the upper left, while most of the young samples are located in the lower right part of the plot. See methods section that demonstrates samples \u003cem\u003es525_F_Yo\u003c/em\u003e and \u003cem\u003es326_M_Fr\u003c/em\u003e \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eas\u003c/span\u003e extreme outliers.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eIdentification of Outlier Samples based on k-means\u003c/h3\u003e\n\u003cp\u003eAfter implementing k-means clustering, samples that do not cluster with their phenotype are potential outliers. After removal of outlying samples, the clustering algorithm is applied again to the remaining samples to check whether all the samples cluster only within their phenotype, and the clustering is repeated until all the samples only cluster within their phenotype. Reducing the samples to two groups of genotypically distinct samples that represent phenotypically distinct groups is of interest to the author on the outcome of disease prediction.\u003c/p\u003e\u003cp\u003eA small experiment is run to show the importance of using a MDS plot to remove samples that evidently seem distant from the rest of samples of their phenotype, or samples that clearly mingle within the contrast phenotype. It is shown that the described samples can significantly shift the centroid of one cluster, leading to several samples not being categorized in their parent group.\u003c/p\u003e\u003cp\u003eThe samples that are removed after the repetitive clustering that leaves only samples that cluster with their phenotype are: \u003cem\u003es525_F_Yo\u003c/em\u003e, \u003cem\u003es514_F_Yo\u003c/em\u003e, \u003cem\u003es508_M_Yo\u003c/em\u003e and \u003cem\u003es530_F_Yo\u003c/em\u003e among young samples; and \u003cem\u003es326_M_Fr\u003c/em\u003e, \u003cem\u003es323_M_Fr\u003c/em\u003e, \u003cem\u003es315_M_Fr\u003c/em\u003e, \u003cem\u003es317_M_Fr\u003c/em\u003e, \u003cem\u003es318_F_Fr\u003c/em\u003e, \u003cem\u003es302_M_Fr\u003c/em\u003e, \u003cem\u003es303_M_Fr\u003c/em\u003e, and \u003cem\u003es306_M_Fr\u003c/em\u003e among frail samples. With the remaining surviving samples, in all 12 runs of the k-means algorithm, all the phenotypically young samples cluster in one category, and all the phenotypically frail samples cluster in one category! The dataset of 22 young samples and 18 frail samples are deemed robust in their phenotypic and genotypic classifications, and are thus used to train and validate the gene selection. The selected samples are split into a training group made of 11 young and 11 frail samples, and a validation group made up of 11 young and 7 validation samples.\u003c/p\u003e\u003cp\u003eAs a remarkable expectation, all the samples that were visible outliers from MDS plots are among the samples that did not survive the clustering into two phenotypically distinct groups.\u003c/p\u003e\n\u003ch3\u003eSelection of Gene Features\u003c/h3\u003e\n\u003cp\u003eThe ReliefF algorithm is a feature selection method that can effectively eliminate irrelevant features, by removing features with weights that are below a given threshold \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W\\)\u003c/span\u003e\u003c/span\u003e, leaving a feature subset that is relevant to distinguishing different types (or groups) of samples [5]. Features here are genes. The weight of a feature is determined by its ability to distinguish short-distance samples of contrasting phenotypes. A randomly chosen number of samples \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:m\\)\u003c/span\u003e\u003c/span\u003e is used to perform selection, and for each sample, there is a given number (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e) of nearest samples hits and misses to be considered. Although \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k=10\\)\u003c/span\u003e\u003c/span\u003e is used in many investigations as the most common assumption [10], a value for \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e lesser than that is used. With up to 11 training samples per phenotype, the maximum value for \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e can only be 10. Adopting the maximum value does not allow for a significant variation in the nearest neighbor samples for each target sample. The author is interested in the outcome of this variation on classification. The implemented scheme here is as in Ref. [5]. The algorithm of implementation is described in the methods section, and is coded in RStudio. The selected relevant genes are ranked starting from the most relevant to the least relevant. The feature selection does not take care of redundancy among the features, and so a correlation algorithm has to be applied in order to eliminate redundancy.\u003c/p\u003e\u003cp\u003eThe relevant genes are provided as input to the distance correlation algorithm, in order to remove genes that are redundant, so that the selected genes weakly depend on one another. The counts of a gene across samples constitute a vector of numbers. The aim in distance correlation is to determine the level of dependence of two vectors, denoted as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:R\\)\u003c/span\u003e\u003c/span\u003e. Distance correlation satisfies \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:0\\:\\le\\:R\\:\\le\\:1,\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:R=0\\)\u003c/span\u003e\u003c/span\u003e only if vectors \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:X\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Y\\)\u003c/span\u003e\u003c/span\u003e are independent [\u003ca class=\"FNLink\" href=\"#Fn12\" id=\"#FNLinkFn12\"\u003e\u003c/a\u003e]. The distance correlation is evaluated by the function dcor(\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:X,Y\\)\u003c/span\u003e\u003c/span\u003e) which is provided by the \u003cem\u003eenergy\u003c/em\u003e package in RStudio. Genes are determined to be correlated beyond a threshold dependence limit \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{R}_{th}\\)\u003c/span\u003e\u003c/span\u003e. Distance correlation is preferred here over other correlation schemes such as Pearson correlation due to its ability to capture both linear and non-linear relationships between vectors, as discussed in Refs. [\u003ca class=\"FNLink\" href=\"#Fn13\" id=\"#FNLinkFn13\"\u003e\u003c/a\u003e, \u003ca class=\"FNLink\" href=\"#Fn14\" id=\"#FNLinkFn14\"\u003e\u003c/a\u003e].\u003c/p\u003e\u003cp\u003eThe cut-off values, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{R}_{th}\\)\u003c/span\u003e\u003c/span\u003e, that define the selected set of classifier genes are determined based on the number of classifier genes desired. The aim is to end up with a small set of classifier genes, ~\u0026thinsp;10, that still closely reproduces the classification tendencies in validation and test samples.\u003c/p\u003e\u003cp\u003eThe correlation between two genes is tested over frail samples only. The author projects that including both samples of contrasting phenotypes will provide a sharp change in the correlation behavior at the transition between samples of contrasting phenotypes. Such behavior will disrupt the true correlation of genes among samples of the phenotype that characterizes the condition of interest \u0026ndash; in this case, frailty. A number of relevant genes survive as uncorrelated after running several iterations in which a surviving gene is tested for correlation against all other genes of lower weight in relevance to frailty. The select uncorrelated genes can be called classifier genes, as it is adequate to classify for frailty based on them.\u003c/p\u003e\u003cp\u003eIt is demonstrated in the following sub-sections that classifier genes that are obtained from the surviving MDS plot samples are adequate in classifying test samples, with a score of \u003cb\u003e5/7\u003c/b\u003e for young validation samples, and a score of \u003cb\u003e6/7\u003c/b\u003e for frail validation samples, based on kNN classification. A slightly better accuracy is obtained with classifier genes obtained based on surviving k-means samples, with a score of \u003cb\u003e10/11\u003c/b\u003e for young validation samples, and a score of \u003cb\u003e5/7\u003c/b\u003e for frail validation samples.\u003c/p\u003e\u003cp\u003e\u003cb\u003eComputation with samples (22 in each phenotypic category) selected from visualization of MDS plots at\u003c/b\u003e \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{\\delta\\:}\\varvec{W}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e= 0.01\u003c/b\u003e\u003c/p\u003e\u003cp\u003eIn the selected dataset that is made of 22 samples in each category of frail or young samples after removal of outlier samples evident from the MDS plot, 58238 genes survive after duplicate ensemblID genes are removed from the original 58283 genes. After stripping off genes that have a total count less than 10 across all samples, 24591 genes are remaining. Any gene outlier counts (67 counts) are replaced prior to normalization. Annotation of the genes is performed to find 24253 genes in the biomaRt database. And after taking off duplicate gene names, the number of surviving genes is 20638.\u003c/p\u003e\u003cp\u003eThe samples are split into 30 training samples (15 in each phenotype) and 14 validation samples. The parameters that define the selection of relevant features are \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:m=20\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k=9\\)\u003c/span\u003e\u003c/span\u003e, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W\\)\u003c/span\u003e\u003c/span\u003e = 0.01. 10474 genes prevail as relevant to frailty.\u003c/p\u003e\u003cp\u003eIn distance correlation, \u003cb\u003e37\u003c/b\u003e genes prevail as weakly correlated features at \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{R}_{th}=0.4\\)\u003c/span\u003e\u003c/span\u003e. Performing k-means for the training and validation samples, based on the \u003cb\u003e37\u003c/b\u003e classifier genes, consistently, 1/22 frail sample (\u003cem\u003es306_M_Fr\u003c/em\u003e) only clusters among all young samples, while 3/22 young samples (\u003cem\u003es512_M_Yo\u003c/em\u003e, \u003cem\u003es527_M_Yo\u003c/em\u003e, and \u003cem\u003es528_M_Yo\u003c/em\u003e) cluster among all frail samples. The representation provided by the classifier genes seems quite satisfactory. The number of iterations required to attain convergence is typically less than 5 in all 12 independent runs. The graph indicating the cluster of the samples based on the \u003cb\u003e37\u003c/b\u003e classifier genes only is depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;2\u003c/b\u003e, and a similar graph for the entire set of \u003cb\u003e20638\u003c/b\u003e genes with normalized counts is displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e. The sample misfits (not clustering with the rest of their phenotype) for the clustering based on the entire gene-set are \u003cem\u003es302_M_Fr\u003c/em\u003e, \u003cem\u003es303_M_Fr\u003c/em\u003e, \u003cem\u003es306_M_Fr\u003c/em\u003e, and \u003cem\u003es318_F_Fr\u003c/em\u003e.\u003c/p\u003e\u003cp\u003eWhen looking at sample \u003cem\u003emisfits\u003c/em\u003e \u0026ndash; samples that do not cluster with the bulk of their phenotype, there are 4 \u003cem\u003emisfits\u003c/em\u003e out of 44 samples in the clustering that is based on the entire gene-set, and also 4 \u003cem\u003emisfits\u003c/em\u003e in the clustering that is based on the classifier genes only. There is one common \u003cem\u003emisfit\u003c/em\u003e in either dataset \u0026ndash; \u003cem\u003es306_M_Fr\u003c/em\u003e, resulting in 3 uncommon misfits. Based on k-means, the classification score of the classifier genes relative to the entire gene-set can be obtained as in \u003cb\u003eEq.\u0026nbsp;(1)\u003c/b\u003e, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{n}_{mis}\\)\u003c/span\u003e\u003c/span\u003e is the number of misfits that are dissimilar to both datasets, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{samples}\\)\u003c/span\u003e\u003c/span\u003e is the number of dataset samples.\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\left(1-\\:{n}_{mis}/{N}_{samples}\\right)\\times\\:100\\%\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(1\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe percentage classification score of the classifier genes relative to that of the entire gene-set, after applying \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{n}_{mis}\\)\u003c/span\u003e\u003c/span\u003e = 3, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{samples}\\)\u003c/span\u003e\u003c/span\u003e = 40, in Eq.\u0026nbsp;(1) is thus \u003cb\u003e92.5%\u003c/b\u003e.\u003c/p\u003e\u003cp\u003eA validation dataset that is composed of the uncorrelated (classifier) genes only in the validation samples is produced. Each validation sample is tested for frailty by calculating the Euclidean distance between the validation sample and an equal number of training frail and young samples that are randomly chosen. The validation sample is classified into the phenotype of the sample that is closest in Euclidean distance to it. In the present scenario, all the training samples (30) are used to test the validation samples. Classification results show that the same samples that did not belong to their parent groups in k-means clustering of training and validation samples do not also classify by their phenotype. This is a remarkable coincidence in results of the supervised learning approach, \u003cem\u003ekNN\u003c/em\u003e classification, and the unsupervised learning approach, k-means. The validation samples are: \u003cem\u003es527_M_Yo\u003c/em\u003e, \u003cem\u003es528_M_Yo\u003c/em\u003e, and \u003cem\u003es306_M_Fr\u003c/em\u003e. This means 5/7 young validation samples classify by their phenotype, and 6/7 frail validation samples classify by their phenotype. For each validation sample, a table of training samples and their distance from the validation sample are produced.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eComputation with samples only that survive in their phenotypic group in k-means clustering (22 young samples and 18 frail samples) at\u003c/b\u003e \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{\\delta\\:}\\varvec{W}=0.01\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp\u003eIn the selected dataset of 22 young samples and 18 frail samples that survive after maintaining only samples that consistently cluster in their phenotypic group in k-means clustering, 58238 genes survive after genes that have duplicate \u0026ldquo;ensembl ID\u0026rdquo; names are removed from the original 58283 genes. After stripping off genes that have a total count less than 10 across all samples, 24238 genes are remaining.\u003c/p\u003e\u003cp\u003eThe normalized dataset is submitted to k-means clustering. In all 12 independent k-means runs, all the frail and young samples are grouped separately, and in typically 3\u0026ndash;4 iterations. The clustering of the frail and young samples into separate categories is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e4\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e. Annotation of the genes is performed to find 23905 genes in the biomaRt database. And after taking off genes with duplicate gene names, the number of surviving genes is 20411.\u003c/p\u003e\u003cp\u003eThe 40 samples are split into 22 training samples (11 in each phenotype) and 18 validation samples (11 young and 7 frail). Nearest misses and hits for each training sample are found. 18 random samples out of 22, and 7 nearest sample hits and misses, are used to perform selection of relevant genes to frailty based on the criterion \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W=0.01\\)\u003c/span\u003e\u003c/span\u003e. 9620 genes have prevailed as relevant to frailty. In distance correlation across samples of the frail category, \u003cb\u003e54 genes\u003c/b\u003e prevail as weakly correlated features at \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:R\\:\\le\\:0.4\\)\u003c/span\u003e\u003c/span\u003e. Using the dataset that is inclusive of training and validated samples with classifier genes only, k-means converges to classify all 18 frail samples in one cluster, and all 22 young samples except \u003cem\u003es516_F_Yo\u003c/em\u003e and \u003cem\u003es522_F_Yo\u003c/em\u003e in one cluster, consistently in 12 independent runs. The described outcome of the k-means clustering of the frail and young samples is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e4\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;2\u003c/b\u003e. Sample \u003cem\u003es516_F_Yo\u003c/em\u003e is a validation sample and sample \u003cem\u003es522_F_Yo\u003c/em\u003e is a training sample. The classification score of classifier genes relative to the classification of the entire gene-set, based on Eq.\u0026nbsp;(1), is \u003cb\u003e95%\u003c/b\u003e.\u003c/p\u003e\u003cp\u003eA validation dataset that is composed of the classifier genes only for the validation sample is produced. Each validation sample is tested for frailty by calculating the Euclidean distance between the validation sample and an equal number of training frail and young samples that are randomly chosen. In the present scenario, all the training samples (22) are used to test the validation samples. For each validation sample, a table of training samples and their distance from the validation sample is produced. Classifying a sample by the phenotype of the closest sample in the table: all the 11 young validation samples classify as young except \u003cem\u003es516_F_Yo\u003c/em\u003e; among the 7 frail validation samples, only samples \u003cem\u003es322_F_Fr\u003c/em\u003e and \u003cem\u003es328_F_Fr\u003c/em\u003e have a young sample as nearest sample. However, when the average distance of a validation sample from frail samples is compared to the average distance from young samples, all frail samples are validated as frail, and all young samples are validated as young, equivalent to k-means classification based on the entire gene-set!\u003c/p\u003e\u003cp\u003eClassifying validation samples by the phenotype of the closest-by-distance training sample in the dataset made of samples that have survived outlier detection using k-means, 10/11 young samples and 5/7 frail samples are properly validated.\u003c/p\u003e\u003cp\u003eThe dataset that is composed of samples that survived outlier detection by using k-means is used for further analyses that involve selection of classifier genes and testing of samples for classification in this work. This dataset boasts of slightly better results than the dataset with samples that have survived MDS plot visualization, although classification scores with either dataset are very satisfactory \u0026ndash; above 90%. The excellent scores have been obtained with classifier genes that are above 30 in number. An outcome with only about 10 classifier genes is of interest to the author. This requires tightening the feature selection criteria. In processing attempts to obtain\u0026thinsp;~\u0026thinsp;10 classifier genes, it is noticed that most of the surviving genes are those with very low counts in the training samples \u0026ndash; less than 1 count per sample on average. Typically for these genes, in each phenotypic group, there are more than 4 samples with zero counts. The noted genes have a relevance weight well above the threshold of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W=\\:\\)\u003c/span\u003e\u003c/span\u003e0.01, and they show very weak correlation, leading to their survival as classifier genes. The sparse and very low gene count in a few samples only can be viewed as noise. Despite the noise, these genes survive as relevant to frailty and most of them would weakly correlate among them to further survive as classifier genes. To avoid the chancy scenario of genes with noisy counts surviving, it is best to take them off, based on a given criterion: genes with a total normalized count across the training samples that is less than a given fraction (here 20/22) of the number of training samples are removed. Prior to selection of relevant genes with the dataset of 22 training samples, any annotated genes with a total count less than 20 across all the samples are removed. This reduces the gene set from 20411 to 17222 genes.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eClassification of Samples (24 young and 18 frail) based on Genes that have survived noise removal\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe parameters that define the selection of relevant features are \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:m=18\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k=7\\)\u003c/span\u003e\u003c/span\u003e, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W\\)\u003c/span\u003e\u003c/span\u003e = 0.03. It is worth noting that the surviving set of relevant genes and their weighted order of relevance depends on the number of nearest hits and misses, as well as the random samples chosen. This affects the eventual selection of classifier genes. It is worth considering making a series of runs in which the choice of random samples is different in each run, resulting in different sets of classifier genes. It would be worth confirming that the different sets of classifier genes are reasonably equivalent by producing similar patterns in classification. The most recurrent and most relevant genes across the different sets of classifier genes could become a final choice set.\u003c/p\u003e\u003cp\u003eA number of independent runs in which different random samples are used to determine the relevant genes are discussed. The same selection criterion for non-correlation, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:R\u0026lt;\\:{R}_{th}\\)\u003c/span\u003e\u003c/span\u003e, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{R}_{th}\\)\u003c/span\u003e\u003c/span\u003e= 0.5, is applied in all runs. After obtention of the classifier genes, the datasets of training and validation samples that are composed of the classifier genes only are submitted to k-means clustering, in order to verify how well the samples become classified into 2 separate clusters, based on the classifier genes only. It has been shown earlier in this work that the outcomes of k-means and kNN classification are quite similar and coincident in some datasets; k-means is chosen here for its relative simplicity in report of the results. The results in 5 cases (\u003cb\u003eA\u003c/b\u003e, \u003cb\u003eB\u003c/b\u003e, \u003cb\u003eC\u003c/b\u003e, \u003cb\u003eD\u003c/b\u003e, \u003cb\u003eE\u003c/b\u003e) of the independent runs are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The sample misfits shown do not cluster with the rest of their phenotype in all the k-means runs. If the sample does not cluster in a significant number of the runs (more than a quarter), it is also defined as a misfit. Such is the case for sample \u003cem\u003es517_F_Yo\u003c/em\u003e in case D. In each k-means run, the number of iterations required for convergence is in the range of ~ 2\u0026ndash;5. The number of classifier genes obtained in the independent selection runs is in the range 9\u0026ndash;12. A classification score for the classifier genes relative to the entire gene-set is above 90%, except in case A, where the score is 82.5%. Interesting finding: Based on these classifier genes, the few samples that do not regularly cluster with their phenotype are samples of female gender! This may be explained by the tendency to exercise less in females, resulting in their relative disposition to frailty.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eOutcome of gene selection and k-means classification in 5 independent selection runs\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCase\u003c/p\u003e\u003cp\u003e(# of relevant genes)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e# of classifier genes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eClassifier Genes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSample misfits\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003escore\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eA\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(5791)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSEMA6A, FAM153A, CRADD-AS1, VWC2L, LRRC3B-AS1, CYP3A43, MIR5697, FAM178B, CFAP107, RPS2P32, NACA2, NUDT12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003es301_M_Fr\u003c/em\u003e, \u003cem\u003es305_M_Fr\u003c/em\u003e, \u003cem\u003es316_F_Fr\u003c/em\u003e, \u003cem\u003es322_F_Fr\u003c/em\u003e, \u003cem\u003es507_F_Yo\u003c/em\u003e, \u003cem\u003es517_F_Yo\u003c/em\u003e, \u003cem\u003es518_M_Yo\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e82.5%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eB\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(6083)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eC10orf71-AS1, RANBP17, CRADD-AS1, CDC26P1, RTL9, CYP3A43, MIR3682, RPS2P32, CFAP107, RNA5SP466\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003es516_F_Yo\u003c/em\u003e, \u003cem\u003es522_F_Yo\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e95%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eC\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(5858)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eC10orf71-AS1, CRADD-ASI, CDC26P1, GSG1L, CYP3A43, LRRC3B-AS, RPS2P32, CXCL3, FAM238B, CFAP107\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003es516_F_Yo\u003c/em\u003e, \u003cem\u003es522_F_Yo\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e95%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eD\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(5697)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eC10orf71-AS1, CRADD-AS1, CYP3A43, CDC26P1, GSG1L, RANBP17, RTL9, RPS2P32, CFAP107\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003es516_F_Yo\u003c/em\u003e, \u003cem\u003es517_F_Yo\u003c/em\u003e, \u003cem\u003es522_F_Yo\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e92.5%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eE\u003c/b\u003e\u003c/p\u003e\u003cp\u003e(3861)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eC10orf71-AS1, CDC26P1, RTL9, CRADD-AS1, LINC01121, CYP3A43, RANBP17, RPS2P32, SHISA8, RNA5SP466\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003es516_F_Yo\u003c/em\u003e, \u003cem\u003es522_F_Yo\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e95%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\n\u003ch3\u003eMaking Use of the determined Classifier Genes\u003c/h3\u003e\n\u003cp\u003eOnce the classifier genes are determined, it is of interest to test the frailty condition for an arbitrary sample similar to the approach discussed in the classification of validation samples. Given that the rna-seq counts for the test sample will be obtained from an experimental set-up different from the set-up in which the rna-seq counts of the training samples were obtained, \u003cem\u003ebatch effects\u003c/em\u003e will have to be accounted for in order to determine the rightful Euclidean distance of the test sample from the training samples. The variation on the handling of biological samples in different rna-seq quantitation experiments influences the number of counts recorded for RNA transcripts. As we aim to test for the classifier genes in datasets that have been characterized for frailty in other experiments, with lack of knowledge of batch effects on the rna-seq counts in the separate experiments, we resort to clustering of the test samples based on the classifier genes. Separation of the test samples into their phenotypically characterized groups based on the classifier genes alone will be a strong confirmation of the authenticity of the classifier genes. Again, k-means clustering is used for this endeavor.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eAnalysis of Test Dataset\u003c/h2\u003e\u003cp\u003eThe classifier genes are tested across a GEO dataset of peripheral blood mononuclear cells (PBMC) genes in 16 middle aged individuals that have been tested for frailty. Although the tissue from which the genes of the test dataset are derived is not the same tissue from which the classifier genes are derived, it is of interest to reveal the tendency to identify frail samples from non-frail samples.\u003c/p\u003e\u003cp\u003eThe table of 39374 normalized and annotated RNA-seq gene counts data composed of 16 samples is read into RStudio 4.3. There are no duplicate gene names. A MDS plot of the 16 samples based on the original gene-set is displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e. There is no clear-cut division between distinct frail and non-frail regions, but there are regions of space which are dominated either by frail samples or by non-frail samples. Examples of identified regions can be named. Samples frail_1, frail_2, frail_4 are seen to form a contiguous region in which there is no intervening sample of the contrast phenotype. Other such regions are: samples non-frail_1, non-frail_2, non-frail_5; and samples non-frail_6, non-frail_7 and non-frail-8; and samples frail_3, frail_6, frail_7, frail_8.\u003c/p\u003e\u003cp\u003eThe results of k-means clustering, displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;2\u003c/b\u003e, consistently show that the following groups of samples belong to one cluster indexed as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:1\\)\u003c/span\u003e\u003c/span\u003e: frail_1, frail_2, frail_4; non-frail_1, non-frail_2, non-frail_3; and non-frail_5, non-frail_6, non-frail_8. All these samples are also located on the left side of the MDS plot, at leading logFC dim1\u0026thinsp;\u0026lt;\u0026thinsp;0, of Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e. The remainder of samples belong to the cluster indexed as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:2\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eFor k-means to display results that provide observations similar to those noted from visualization of the MDS plot, a multi-cluster k-means approach should be adopted. The purpose is to identify clusters that show similar sample compositions as the noted regions in the MDS plot. The number of clusters adopted is four (4).\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eClassification of Test Dataset Samples based on k-means with 4 clusters\u003c/h3\u003e\n\u003cp\u003e4-centroid k-means with random initialization is applied to the entire gene-set of the test dataset. 5 out of 6 samples in cluster indexed as 4 are frail samples \u0026ndash; frail_3, frail_5, frail_6, frail_7, frail_8; non-frail_4 is the only non-frail sample. In the cluster indexed as 3, there are 2 frail samples (frail-1 and frail-2) and 1 non-frail sample (non-frail-3). In cluster 2, there are 3 non-frail samples (non-frail_1, non-frail_5, and non-frail_6) and 1 frail sample (frail_4). And finally, cluster 1 has only non-frail samples \u0026ndash; non-frail_2, non-frail_7, non-frail_8. [See \u003cb\u003eFig.\u0026nbsp;6-a\u003c/b\u003e for display of the samples in clusters]. The k-means clustering reflects the proximity of samples in the MDS plot, and most samples cluster with samples of their own phenotype. Samples of a minority phenotype in a cluster are called \u003cem\u003emisfits\u003c/em\u003e, similar to the earlier description provided of samples that do not cluster with the rest of their phenotype. There are thus 3 misfits upon clustering with the entire gene-set, which are: non-frail_4, non-frail_3, and frail_4.\u003c/p\u003e\u003cp\u003ek-means is then applied with 4 centroids to classify the test dataset samples based on classifier genes. The classification of samples closely matches that obtained with the entire gene-set, validating the adequacy of the classifier genes. In most scenarios, the 3 frail samples \u0026ndash; frail_1, frail_2, and frail_4 \u0026ndash; are classified in a cluster separate from the remaining frail samples, while the frail samples frail_6, frail_7, and frail_8 are always in one cluster, to closely depict what has been witnessed as sub-regions of a unique phenotype in MDS. The clustering meets the expectations of classification in which the clusters are having the majority of samples as one phenotype. k-means is performed with the different sets of classifier genes obtained in the previously described cases \u003cb\u003eA\u003c/b\u003e-\u003cb\u003eE\u003c/b\u003e.\u003c/p\u003e\n\u003ch3\u003ek-means Classification of Test Samples based on defined sets A-E of classifier genes\u003c/h3\u003e\n\u003cp\u003eThe classifier genes obtained from a given dataset may not all be present in a test dataset with samples obtained from a different experiment. If the derived classifier gene is absent from the test dataset or has a zero count in all the test samples, it is replaced by the correlating gene that is closest by relevance to frailty in the training dataset. Among the classifier genes, the top gene C10orf71-AS1 is present but not expressed (zero count) across all samples in the test dataset. To enable the described replacement of genes, tables of the top five (5) correlating genes that are closest by relevance to the classifier genes are produced and used; see Tables\u0026nbsp;\u003cspan refid=\"Tab11\" class=\"InternalRef\"\u003e6\u003c/span\u003e in the methods section. In case B, C10orf71-AS1 is replaced by \u0026ldquo;STRAD8\u0026rdquo;, CDC26P1 is absent and is thus replaced by \u0026ldquo;GSTZ1\u0026rdquo;, RNA5SP466 is absent and is thus replaced by \u0026ldquo;ANKLE1\u0026rdquo;; in case C, C10orf71-AS1 is replaced by \u0026ldquo;TEX56P\u0026rdquo;, CDC26P1 is replaced by SMIM10L1 ; in case D, C10orf71-AS1 is replaced by \u0026ldquo;UNC13C\u0026rdquo;, CDD26P1 is replaced by CD27-AS1; and in case E, C10orf71-AS1 is replaced by \u0026ldquo;OIP5-AS1\u0026rdquo;, CDC26P1 is replaced by C1orf21, RNA5SP466 is replaced by CFAP20. The outcomes of classification of the test samples based on the sets of classifier genes A-E are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Diagrams of clusters containing the test samples are depicted in \u003cb\u003eFig.\u0026nbsp;6b\u003c/b\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e7\u003c/span\u003ea-d. The clustering outcomes are consistent within 4 independent runs, and in each run, convergence occurs in less than 5 iterations. Like in the evaluation for training and validation samples, the classification score is evaluated relative to the outcome of classification that is based on the entire gene-set, such that only dissimilar sample misfits in both outcomes account for the score. The number of dissimilar misfits to the classifier set outcome and the entire gene-set outcome is indicated in the 3rd column of Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Classification scores are \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\ge\\:\\)\u003c/span\u003e\u003c/span\u003e 80% in all 5 cases, with the lowest at 80%, and the highest at 93.75%. In all cases, highlighting samples that are recurrently misfits, the sample non-frail_4 is a misfit in all 5 cases, sample non-frail_5 is a misfit in 3 cases, sample non-frail_3 is a misfit in 2 cases, and sample non-frail_8 is a misfit in 2 cases. All the noted samples, except non-frail_8, are the only misfits in the classification with the entire gene-set, as this shows how the classifier genes closely represent the entire gene-set in classification. Notably, sample non-frail_4 can be seen in the MDS plot of Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e to be significantly closer to frail samples than to non-frail samples. The k-means results can be qualified as robust!\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eOutcome of classification of test dataset samples.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCase\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSample misfits\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e# of dissimilar misfits\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003escore\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eA\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003enon-frail_4, frail_4, frail_5, non-frail_3, non-frail_5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e87.5%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eB\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003enon-frail_4, non-frail_1, non-frail_8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e87.5%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eC\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003enon-frail_4, non-frail_2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e93.75%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eD\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003enon-frail_4, non-frail_8, non-frail_3, non-frail_5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e87.5%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eE\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003enon-frail_4, non-frail_8, non-frail_5, frail_5, frail_1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e80%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eDiscussion of Top Frailty Genes\u003c/h2\u003e\u003cp\u003eA survey is done to see how the relevant frailty genes compare to notable frailty (or aging) genes in the literature. Notable aging genes are drawn from the literature for comparison with the genes that are relevant to frailty in this work. Investigations conducted in Ref. [\u003ca class=\"FNLink\" href=\"#Fn15\" id=\"#FNLinkFn15\"\u003e\u003c/a\u003e] resulted in the identification of top 20 genes that are upregulated with aging, and top 20 genes that are downregulated with aging. The study was done with 53 healthy individuals (22\u0026ndash;83 years old) on 57,205 protein-coding and non-coding RNAs. The healthy condition of the individuals provides an interpretation of gene expression results that is free from the impact of disease. The RNA expressions of the identified genes linearly increase (for upregulated genes) or decrease (for downregulated genes) with age in linear regression plots of the cited Ref. The ranks of the identified notable genes among genes relevant to frailty in this work are determined. The gene selection outcome from the dataset of samples that survived outlier removal by k-means clustering with relevance threshold \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W\\)\u003c/span\u003e\u003c/span\u003e = 0.01 is used as a reference for sourcing the relevance rank. The rank and weight of every notable gene is recorded in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e: the genes that are upregulated with age are tabulated in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e, and those that are downregulated with age are tabulated in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;2\u003c/b\u003e. Out of the top 20 upregulated genes, one of them is without a gene name, leaving 19 genes in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e. Out of the top 20 downregulated genes, 13 of the genes have gene names, as that is the number of genes in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;2\u003c/b\u003e. Genes with \u0026ldquo;NA\u0026rdquo; in the rank and weight are those that are not found among the relevant genes \u0026ndash; there are six such genes: TET2, EFEMP1, FAM49A, SMIM2, MYLPF, and MINOS1. Only TET2 and EFEMP1 are among the annotated genes of the training dataset, as their weight of relevance to frailty is below the accepted threshold of 0.01. The remaining four genes are completely absent from the training dataset of annotated genes. Among the 32 notable aging genes tabulated, 11 genes are among the top 10% rank of relevant frailty genes, as this seems a remarkable representation of the notable genes among the genes that have been determined to be relevant to frailty.\u003c/p\u003e\u003cp\u003eLooking at any set of classifier genes, it may be difficult to identify among them notable frailty genes from the literature. It suffices to show, however, that the classifier genes correlate with the identified notable genes. Notable aging genes that correlate with the top 12 classifier genes are found, enabling thus to witness the link between the classifier genes and the notable aging genes. This correlation is tabulated in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e4\u003c/span\u003e. As a remarkable feat, 26 of the 32 notable aging genes of Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e3\u003c/span\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e correlate with the 12 classifier genes, even if none of the notable genes is among the given classifier set. It can also be observed that the more highly ranked classifier genes correlate with a bigger number of notable frailty genes, as 7 notable frailty genes correlate with the top classifier. The classifier genes that correlate with several notable genes may not have a fully or appropriately defined role in frailty in the literature. This insufficiency provides grounds for determining or investigating the roles of such classifier genes!\u003c/p\u003e\u003cp\u003eThe top 500 relevant genes are submitted to the Molecular Signatures Database (MSigDB) \u0026ndash; a multicollection of databases among which are curated gene sets and databases of ontology gene sets, in order to find gene sets that overlap with the submitted genes. The online search for the 20 most populated gene sets is done for the 5 variants of relevance gene sets A-E, and the overlapping genes for every gene set are described. The results showing the gene-set names, the overlapping genes and their description are provided in files with the \u0026lsquo;tsv\u0026rsquo; extension in the supplementary section. The files are named as overlap_A.tsv, overlap_B.tsv, overlap_C.tsv, overlap_D.tsv, and overlap_E.tsv, for the 5 different variants A-E, respectively. Although searches based on the different variants display different combinations of gene sets, there are gene sets that are recurrent across the different combinations. Such gene sets and some of their constituent genes are tabulated in Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e5\u003c/span\u003e. In Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e5\u003c/span\u003e, are also indicated gene sets that are of characteristic evidence to frailty, e.g. the gait disturbance gene set. The gene-set names in Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e5\u003c/span\u003e are indicative of processes and conditions at different levels of organization. Parts of some gene set names at the cellular level are: mitochondrion, apoptosis, target genes, protein modification, nuclear complex, and microRNA. Parts of gene set names at the tissue and organ level are: skeletal muscle cells, fetal lung cells, and pancreas ductal cells. Some of the conditions indicated by the gene set names are: Alzheimer\u0026rsquo;s disease, nasopharyngeal carcinoma, bladder cancer, chronic myelogenous leukemia, and gait disturbance. A metabolic process that is indicated is phosphorus metabolic process.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e. The rank and weight of notable [upregulated with age] aging genes in Ref. [15] among genes that are relevant to frailty.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003egene\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003erank\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:W\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFunctions and characteristics, described in Ref. [15]\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eC12orf75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e125\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.213865\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEnergy metabolism, insulin signaling, skeletal muscle weakness\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCDKN2B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e85\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.226386\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ep15 tumor suppression, aneurysm formation, aging\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSKAP2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.322864\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSarcomere function and regulation\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCFAP61\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e72\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.233005\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eHigh expression in old skeletal muscle, skeletal muscle weakness\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePLAG1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e96\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.221592\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBovine fetal development and body size\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLGI1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e119\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.214702\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSkeletal muscle weakness, limbic encephalitis, and epilepsy\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIRS2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e701\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.13417\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eInsulin signaling, lipid metabolism in skeletal muscle, mice longevity\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFEZ2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e233\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.188437\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003efasciculation, axonal elongation, skeletal muscle aging\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePFKFB2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e412\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.159812\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eHeart isozyme\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eESRRG\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7844\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.01944\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eType I muscle fiber development, estrogen signaling, obesity susceptibility\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFAM83B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1737\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.09142\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eHypoxia response pathway, aging, skeletal muscle weakness\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFAM117B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e8468\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.015853\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSarcoidosis susceptibility, candidate for early onset myocardial infarction\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRIM1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3191\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.061272\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSkeletal muscle aging, smooth muscle contractility, CNS development\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNR2F2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2918\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.065517\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eMyogenesis, skeletal muscle development, aging, muscular dystrophy\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLPP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e4551\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.044722\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSmooth muscle differentiation, cell migration and proliferation\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTET2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eKCNQ5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1847\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.08812\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePotassium channel subunit\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEFEMP1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eENAM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e8813\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.01402\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTooth enamel formation\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;2\u003c/b\u003e. The rank and weight of notable [downregulated with age] aging genes in Ref. [15] among genes that are relevant to frailty.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003egene\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRank\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:W\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFunctions and characteristics, described in Ref. [15]\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMYLK4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e344\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.16906\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eweakness, contraction, motility, TGF-beta pathway, circadian rhythms\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMYL1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3996\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.050828\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eExpressed in fast-type muscle\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLDHA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e4925\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.040896\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePorcine energy regulation, glycogen metabolism, and glycolysis of skeletal muscle\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFAM49A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTPM1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1516\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.098513\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eModulates actin-myosin interactions\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTPI1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e712\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.133497\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGlycolysis enzyme, energy generation for muscle cells in chickens\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSLN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1665\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.093632\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eInhibits rat sarcoplasmic Ca2+-ATPases, thermogenesis, muscle performance\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTNNT3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2938\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.065257\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSkeletal muscle contractibility, porcine muscle development, tropomyosin binding\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSMIM2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTUBA8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e8532\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.01553\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eIntracellular transport of macromolecules and cytoskeletal support\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMYLPF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMINOS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNA\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMSS51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3413\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.058304\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSkeletal muscle-specific modulator of cellular metabolism\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eRecord of the notable aging genes from Ref. [15] that correlate with a set of 12 frailty classifier genes.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRank\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eClassifer Gene\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCorrelating notable aging genes in Ref. [15]\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eC10orf71-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eC12orf75, CDKN2B, SKAP2, CFAP61, PLAG1, LG1, FEZ2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e188\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCDC26P1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eIRS2, PFKFB2, MYLK4, TPI1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e704\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCRADD-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNone\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e948\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRANBP17\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTPM1, SLN\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e1544\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCYP3A43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNone\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e1551\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLINC01121\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFAM83B, CRIM1, NR2F2, KCNQ5\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e2757\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRTL9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMYL1, TNNT3, MSS51\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e3190\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSHISA8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLPP, LDHA,\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e5232\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRPS2P32\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNone\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e5709\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRNA5SP466\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTUBA8,\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e6313\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCCDC144NL-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eESRRG, FAM117B,\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e7312\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYBX2P2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eENAM\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSummary Results of MSigDB query of gene sets that overlap with top 500 relevant frailty genes. The first word of every gene set name is the name of the database collection or the source of provision of the gene set.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMSigDB gene set\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDescription of gene set\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e# of overlap genes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eExample genes\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGRAESSMAN APOPTOSIS BY DOXORUBUCIN DN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes downregulated in ME-A cells (breast cancer) undergoing apopotosis in response to doxorubicin\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e65\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSCL44A2, IRF1, FEZ2, URI1, PP1CC, HSPE1, SFXN4, STX17, RFK, SIRT4, KIF5B, CTNNBIP1, HNRNPL, NAGK, GJC1, STRADA, PLP2, ZNF768, TRIM44, PPP1CC, SRSF1, TARS2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBLALOCK ALZHEIMERS DISEASE UP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes up-regulated in brain from patients with Alzheimer\u0026rsquo;s disease\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e59\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSEMA6A, FEZ2, USP36, STX4, ACIN1, NNT, GBE1, PKRCH, STRADB, NCKAP1, ITGA6, ITGA1, TACC1, RGCC, SIRT4, URI1, RFK, CRADD, IGF1, HNRNPK, ERBB2, TRAP1, RTL10, PARP1, CYCS, DAPK3, TNFRS10B, SLIT2, CDKN1A, ADAM8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGOCC MITOCHONDRION\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eA semiautonomous, self-replicating organelle that occurs in varying numbers, shapes, and sizes in the cytoplasm of eukaryotic cells. It is notably the site of tissue respiration\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePDP1, NKTR, GRSF1, PDHX, MRPL38, DHX32, PARP1, FAM210A, MT-ND3, RTL10, SMIM20, URI1, MT-CO2, NNT, MTHFDIL, CHCD3, MFF, MCM7, GOT2, TARS2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePAX3 TARGET GENES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes containing one or more binding sites for Uniprot:P23760 (PAX3) in their promoter regions (TSS-1000,+100 BP) as identified by GTRD version 20.6 ChIP-seq harmonization.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTEX56P, SCYL2, PCDH17, SEMA6A, HNRNPC, TNFRSF10B, MAP4K, ITFG2, SKAP2, UBE21, ITGA6, CTNNA3, TMEM19, FEZ2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMIR561_3P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes predicted to be targets of miRbase v22 microRNA has-miR-561-3p in miRDB v6.0 with MirTarget v4 prediction scores\u0026thinsp;\u0026gt;\u0026thinsp;80 (high confidence targets)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e33\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRANBP3, SLC25A15, PCDH9, C1orf21, POPDC3, CFAP61, URI1, STX17, PPA2, POU2F1, DAZAP1, SPRED1, ZBTB18, HAPSTR1, PHTF2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDODD NASOPHARYNGEAL CARCINOMA UP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes up-regulated in nasopharyngeal carcinoma (NPC) compared to the normal tissue\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e58\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSEMA6A, CBX1, SRSF7, STRAP, MCM3, TIPRL, ANKIB1, SLC16A1, UBE2Z, ITGA1, MAP2K4, NOLC1, KLHL5, PAIP1, SLC16A1, EIF3J, TIMMDC1, TRMT61B, SCYL2, NNT, MFF, SRSF1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNAKAYA PBMC FLUMIST AGE 18 50YO 7DY UP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes up-regulated in peripheral blood mononuclear cells 7d vs 0d in adults (18\u0026ndash;50) after exposure to FluMist, time point 7D.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e61\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGRSF1, SFXN4, MTIF2, ATPAF1, FEZ2, DYM, RPRD1A, EWSR1, GLG1, PPP2CB, COPS5, PPIE, ATP13A1, WRAP73, CEP164, MORC2, RNF170, ARSG, SCOC, TERF2, NNT, CHCHD3, SRSF1, GOT2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHP ABNORMAL FUNDUS MORPHOLOGY\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAbnormal fundus morphology\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e55\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eCDKN2B, SEMA3E, LPAR6, ANKRD11, PEX2, RNF170, DTNBP1, TENM3, SCYL2, COA8, MT-ND4, PDHX, TIMMDC1, SCL25A46, MT-CO2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDIAZ CHRONIC MYELOGENOUS LEUKEMIA UP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes up-regulated in CD34\u0026thinsp;+\u0026thinsp;cells isolated from bone marrow of CML (chronic myelogenous leukemia) patients, compared to those from normal donors.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e56\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTRAK2, NNT, CYCS, MTIF2, PPA2, NIPSNAP2, SCP2, HNRNPC, KPNA4, MAP4K4, PTP4A2, NDEL1, GBE1, PARP1, SLC16A, PAIP1, CCNG1, PLP2, CDC42BPA, VTI1B, PEX2, SART3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTHAKAR PBMC INACTIVATED INFLUENZA AGE 21 30YO RESPONDERS 28DY DN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes down-regulated in peripheral blood mononuclear cells 28d vs 0d in young adults (21\u0026ndash;30) after exposure to inactivated influenza vaccine, time point 28D\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e44\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSTRADA, STX4, USP36, NISCH, NAGK, RBM23, SEC24C, MAN2C1, WDR6, ADAM8, ARFGAP2, RANBP3, ZFHX3, CEBPB, ATP13A1, VPS52, ITFG2, NKTR\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTFEB TARGET GENES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes containing one or more binding sites for UniProt:P19484 (TFEB) in their promoter regions (TSS-1000,+100 BP) as identified by GTRD version 20.06 ChIP-seq harmonization.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e56\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eMFF, LIG1, FA210A, SKAP2, PRKCH, PRRC2C, MAPKAP1, CEP164, KHDRBS1, STRADA, STX4, AC1N1, HNRNPL, PHFT2, HAPSTR1, ANKRD11, SLC22A18, PPP1R3E, GABBR2, DNAH11, STRADB, MFF\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRUBENSTEIN SKELETAL MUSCLE ENDOTHELIAL CELLS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSame as name of gene set\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e20\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNKTR, ITGA6, PTPRB, PHACTR2, F8, AQP1, ITGA1, TACC1, RGCC, FLT1, SPTBN1, MGLL, MYCT1, NEAT1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHE LIM SUN FETAL LUNG C5 LARGE PRE B CB CELL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLarge pre-B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e53\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePCDH17, CLK2, UBXN2A, C12orf75, DDB2, SART3, TELO2, EDC4, RBMX2, CCNG1, DADK3, NFATC2IP, MCM7, GUSB, NDUFC1, PPA2, MCUR1, SCCPDH, MRPL13, TRAP1, MTIF2, PDHX, CTNBIP1, ADK, CBX1, DAZAP1, SFXN4, MRPL15, ENO3, GOT2, MLEC\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGOBP APOPTOTIC PROCESS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eA programmed cell death process which begins when a cell receives an internal (e.g. DNA damage) or external signal (e.g. extracellular death ligand), and proceeds through a series of biochemical events (signaling pathway phase) which trigger an execution phase.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e65\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSEMA6A, PLAG1, CDKN1A, TMEM163, SLC16A1, IGSF3, ADAMTSL4, DPF2, ERBB3, GFOD1, PCSK7, TJAP1, FCRLA, HNRNPC, PPFIBP1, KCNMA1, KLF5, EIF4H, MRPL13, MTIF2, DCLK1, SPRED1, H1-0, HSD3B7, IGF1, ZFHX3, SASH1, SLC16A1, PRCKH, LDLRAD4, C16orf74, STX4, USP36, CDH5, PDP1, LDLRAD4, MAPK6, RIN1, NDUFC1, GREM1, DPF2, RNF152, IP6K2, ERBB3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGOBP REGULATION OF PROTEIN MODIFICATION PROCESS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAny process that regulates the frequency, rate or extent of the covalent alteration of one or more amino acid residues within a protein\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e50\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNNT, CDKN2C, CCNG1, SVIP, UBXN2A, DCUN1D5, STRADB, CDH5, MAPKAP1, STRADA, CDKN1A, IP6K2, GREM1, TNFRS10B, IL6R, EFNA1, ITGA1, NCK1, RGN, UCN, PHIP, PRNP, ERBB2, CCAR2, SLIT2, ACE, HDAC4, TAB2, NELFA, DOCK7, SPTBN4, GPS2, SASH1, ATP2B4, PARD3, TRIM44, NNMT\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSKIL TARGET GENES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes containing one or more binding sites for UniProt:P12757 (SKIL) in their promoter regions (TSS-1000,+100 BP) as identified by GTRD version 20.06 ChIP-seq harmonization.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSEMA6A, CDKN2C, CMC2, TNFRS10B, DCAF5, TARS2, HMBOX1, H1-10, EWSR1, PNPLA7, DDX51, ZXDC, PYGO2, ATP13A1, FUS, HSPE1, CDC42BPA, AKAPBL, RPS27L, CCAR2, CYCS, CLN8, QRICH1, COA8, HECW2, NEAT1, WDR6, RIN1, LDLRAD4, MTHFDIL, PRKCH, MLEC, MCM7, MRPL13, MTIF2, NUP85\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMURARO PANCREAS DUCTAL CELL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSame as gene set name\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e52\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNNT, CBR1, TNFRS10B, ITGA6, ATP2B4, PARD3, DERL1, CDKN2B, SLC44A2, HEG1, GNG12, GALNT1, H3-3B, PPFIBP1, IGSF3, KLF5, PPIC, IFNARI1, KDM5B, KLHL5, PDP1, MAPK6, TRAK2, MAPKAP1, CAST\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGOBP PHOSPHORUS METABOLIC PROCESS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThe chemical reactions and pathways involving the nonmetallic element phosphorus or compounds that contain phosphorus\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e66\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eCDC42BPA, SASH1, PPP1CC, URI1, STRADA, DCLK1, PHIP, SCL44A2, NAGK, ERBB2, LCLAT1, PDP1, PRCKH, TNFRSF10B, MAPK6, SEMA6A, PARP1, DRC3, COA8, DDX51, ATP13A1, TRAK2, PHACTR2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOSMAN BLADDER CANCER\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenes upregulated in blood samples from blood cancer\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e27\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eEWSR1, GALNT1, NNT, TIPRL, ATF71P, APPL1, KDHRBS1, NRIP1, HNRNPL, PTP4A2, TRAK2, NKTR\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHOUNKPE HOUSEKEEPING GENES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eList of 1130 human and mouse housekeeping genes. This list is the overlap of human genes stably expressed across 52 tissues and cell types.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e45\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTRMT112, QRICH1, HSPE1, DCAF5, PPP1CC, STX4, KHDRBS1, CBX1, VTI1B, THRAP3, EIF4H, UBE21, PARP1, MLEC, SART3, CCAR2, DDX51, VPS52, NXF1, KPNA4, SCYL2, HAPSTR1, NTSC2, LDB1, PPP2CB, MMADHC, NFAT2CIP, ATPAF1, RBMX2, PITHD1, H3-3B, DERL1, APPL1, STRAP, ERLEC1, PBDC1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGOCC NUCLEAR PROTEIN CONTAINING COMPLEX\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eA stable assembly of two or more macromolecules, i.e. proteins, nucleic acids, carbohydrates or lipids, in which at least one component is a protein and the constituent parts function together in the nucleus\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e50\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNUPB5, E2F3, RBMX2, MCM7, EXOSC7, CTNNBIP1, MCM5, SART3, INTSI3, MCM3, PPIE, WDR6, HDAC4, NELFA, PYGO2, DOCK7, INO80E, ELOA, CDKN1A, HNRNPC, DPF2, TEAD1, CXXC1, HNRNPM, CBX4, PPP2R1A, DMAP1, NOC2L, KMT2E, NKF1, SMARCD2, INTS5, TERF2, TDRD3, KMT2B, CRY2, ACTR5, MVP\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHP GAIT DISTURBANCE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGait disturbance\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e50\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGBE1, RNF168, PDHX, TELO2, PDP1, WDRB, STRADA, SLC25A46, MORC2, MFF, COA8, CP, FUS, KDM5B, NTSC2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"MATERIALS AND METHODS","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eProcessing of GEO datasets\u003c/h2\u003e\u003cp\u003eThe genes in the input dataset are annotated as gene IDs, with 2 digits after a dot in the ID. The two digits are meant to indicate the version of the annotation technology used, and thus do not impart any biological meaning. The 2 digits and the dot are trimmed off from the gene IDs. Any rows in the data that are bearing duplicate gene IDs are also deleted from the dataset, resulting in 58238 genes from an initial number of 58283.\u003c/p\u003e\u003cp\u003eThe dataset that is used to construct the set of classifier genes is raw counts of RNA-seq data obtained from muscle biopsies of vastus lateralis, deposited in the Gene Expression Omnibus (GEO) database under the title \u0026ldquo;Muscle function in fit and (pre-)frail males and females\u0026rdquo;, made public in Feb 04, 2023, and accessible from the Gene Expression Omnibus (GEO) with accession number GSE144304. The RNA-seq counts were obtained with the Illumina NextSeq 500 (Homo sapiens) platform from samples that were submitted in Jan 27, 2020. The total number of samples in the dataset is 80, with each sample coming from a separate individual. The individuals are phenotypically categorized as young, fit, or frail. Old subjects (\u0026gt;\u0026thinsp;75 years) were screened for frailty using the Fried criteria. (e.g. weight loss, exhaustion, low physical activity, slowness and weakness) [\u003ca class=\"FNLink\" href=\"#Fn16\" id=\"#FNLinkFn16\"\u003e\u003c/a\u003e]. Based on these criteria the old subjects were divided into two groups; fit (no frailty indicators present) and frail (\u0026ge;\u0026thinsp;1 frailty indicator present). The young subjects were individuals 20\u0026ndash;30 years of age with a healthy BMI (20\u0026ndash;25 kg/m\u003csup\u003e2\u003c/sup\u003e). Several exclusion criteria were applied to the young subjects including cardiac failure, COPD, anemia, dementia, cancer, neuromuscular disorders and diabetes mellitus type I and II.\u003c/p\u003e\u003cp\u003e\u003cb\u003eDetection of Outlier Gene Counts and Normalization\u003c/b\u003e. Normalization of genes and gene count outlier detection and replacement is achieved by programming routine functions in the R package \u003cem\u003eDESeq2\u003c/em\u003e. The function DESeqDataSetFromMatrix() is applied on the raw counts data, with the samples of the same phenotypic condition noted in one distinct group, to produce an object named \u0026lsquo;dds\u0026rsquo;. All genes with the sum of counts across all samples less than 10 are removed by applying the function rowSums(count(dds))\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;10. The function DESeq is further applied on \u0026lsquo;dds\u0026rsquo; as DESeq(dds) to obtain an object named ddsDE. DESeq() by default detects and replaces outlier counts with replacement counts when the number of biological sample replicates is greater than or equal to seven (7). For conditions that contain seven or more replicates, \u003cem\u003eDESeq2\u003c/em\u003e replaces the outlier counts with an imputed value, which is the trimmed mean over all samples, scaled by the size factor; and then re-estimates the dispersion, LFCs and \u003cem\u003eP\u003c/em\u003e values for these genes. For a given sample, the median of the ratios of counts to the geometric mean for each gene across all samples is the size factor for that sample. The automatic identification and replacement of outliers, however, leaves the analysis blind to the detection of gene outliers. A procedure to identify the outlier genes can be implemented to allow for removal of the genes. In the ddsDE object, there is a matrix of preserved original counts identified as assays(ddsDE)\u003cspan\u003e$\u003c/span\u003ecounts and a matrix of replacement counts \u0026ndash; assays(ddsDE)\u003cspan\u003e$\u003c/span\u003ereplaceCounts that has a modified count for outlier counts only. The outlier counts are identified as counts that do not equal one another during comparison of counts in the two matrices. Some genes are outliers in multiple samples for the GSE144304 dataset, identifying a total of 60 outlier counts. One (1) gene is an outlier in 3 differing samples, and seven (7) genes are outliers in two (2) differing samples. The number of unique genes with outlier counts is thus fifty-one (51). Those genes that were identified as outliers only had outlier counts in very few of the samples of either the young or frail category. Any sample is only having few outlier gene counts, less than 5 out of a total of about 50 000 counts. As a result, it is deemed reasonable to strip off the gene or modify the outlier count, instead of removing the sample that has it as an outlier. Here the outlier counts are replaced by the recently described procedure. The counts are normalized from ddsDE through the function counts (ddsDE, normalized\u0026thinsp;=\u0026thinsp;TRUE). \u003cem\u003eDESeq2\u003c/em\u003e performs an internal normalization where the geometric mean is calculated for each gene across all samples. For a given sample, the median of the ratios of counts to the described geometric mean is the size factor for that sample. This procedure corrects for library size and RNA composition bias, which can arise for example when only a small number of genes are very highly expressed in one experimental condition but not in the other. A description of the procedure of removal of outliers and normalization of counts is properly documented in Ref. [\u003ca class=\"FNLink\" href=\"#Fn17\" id=\"#FNLinkFn17\"\u003e\u003c/a\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eIdentification of Outliers based on MDS plots\u003c/h2\u003e\u003cp\u003eThe MDS plot for the original dataset of GSE144304 shows that samples \u003cem\u003es326_M_Fr\u003c/em\u003e and \u003cem\u003es525_F_Yo\u003c/em\u003e are outstandingly distant from the remainder of samples (see Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e7\u003c/span\u003e). These two distant samples are taken off, and the MDS plotted again; see Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e8\u003c/span\u003e. In the new MDS plot of 48 samples, sample s514_F_Yo seems significantly distant from the remainder of young samples, and seems closer to frail samples than to young samples. Sample s323_M_Fr seems clustered by young samples. Samples s508_M_Yo and s530_F_Yo seem closer to more frail samples than young samples. The named samples \u0026ndash; s508_M_Yo, s530_F_Yo, and s514_F_Yo, and s323_M_Fr \u0026ndash; are deemed outliers, and are removed, leaving 22 samples in each category of frail or young samples, out of the initial 24 frail samples and 26 young samples.\u003c/p\u003e\u003cp\u003eAs an alternative procedure to demonstrate that there are samples that are either extremely distant from the remainder of samples of their phenotype in the GSE144304 original dataset of 50 samples, or are closer to samples of the contrast phenotype, the density of samples around each sample can be computed to show the number of samples that are within a given Euclidean distance from that sample. This defined distance or radius is defined to be three times the mean value of the matrix values of Euclidean distances between all the samples. The region within the defined radius of a sample can be called the neighborhood of the sample. The density of samples in the neighborhood of each sample is plotted and displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e9\u003c/span\u003e. The number of samples within the neighborhood of each sample is 47, except for samples \u003cem\u003es326_M_Fr\u003c/em\u003e and \u003cem\u003es_525_F_Yo\u003c/em\u003e that have only 1 sample in their neighborhood. And the neighbor of any of these exclusive samples is the other of the exclusive samples, meaning that they are closer to one another than to the rest of all the samples. With the noted extreme outliers removed, the density of samples in the neighborhood of each sample is plotted and displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e10\u003c/span\u003e: in this plot, the radius that defines the neighborhood of a sample is the mean value of the matrix of Euclidean distances for all the remaining 48 samples. Samples \u003cem\u003es314_F_Fr\u003c/em\u003e and \u003cem\u003es514_F_Yo\u003c/em\u003e have the least number of neighbors \u0026ndash; six (6), as they correspond to sample indices 8 and 35, respectively. However, for sample \u003cem\u003es514_F_Yo\u003c/em\u003e, all the six neighbors are frail samples, namely: \"\u003cem\u003es301_M_Fr\u003c/em\u003e\", \"\u003cem\u003es314_F_Fr\u003c/em\u003e\", \"\u003cem\u003es316_F_Fr\u003c/em\u003e\", \"\u003cem\u003es320_F_Fr\u003c/em\u003e\", \"\u003cem\u003es325_F_Fr\u003c/em\u003e\", and \"\u003cem\u003es331_F_Fr\u003c/em\u003e\". And for sample \u003cem\u003es314_F_Fr\u003c/em\u003e, its six neighbors are mostly frail samples, with the exception of \u003cem\u003es514_F_Yo\u003c/em\u003e in that fold; the frail samples are \u003cem\u003es301_M_Fr\u003c/em\u003e, \u003cem\u003es305_M_Fr\u003c/em\u003e, \u003cem\u003es320_F_Fr\u003c/em\u003e, \u003cem\u003es325_F_Fr\u003c/em\u003e, and \u003cem\u003es331_F_Fr\u003c/em\u003e. Sample \u003cem\u003es514_F_Yo\u003c/em\u003e is not only relatively distant from the remainder of samples, but is having only frail samples in its neighborhood. It is thus an outlier. The results of this neighborhood analysis confirm the observations that have been noted from the MDS plots.\u003c/p\u003e\u003cp\u003eThe described procedure by which the samples density around each sample is computed to show the number of samples that are within a given Euclidean distance from that sample is useful for spotting samples that are evidently outliers.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003ek-means clustering\u003c/h2\u003e\u003cp\u003eBy the experience of several investigators on k-means clustering, the initial choice of centroids can determine the outcome of classification. A \u003cem\u003erandom initialization\u003c/em\u003e procedure is adopted here: half of the samples are randomly selected to belong to one cluster, while the other half belongs to another cluster. The centroid of each initial cluster is determined as an imaginary (or virtual) sample of all the genes, with each gene having a count that is the mean of the counts for that gene across samples belonging to that cluster. After the centroids are generated, all the samples are classified according to their closest Euclidean distance to the centroids. A sample is adopted into a cluster if the Euclidean distance of the sample to the cluster centroid is closest among all cluster centroids. The resultant cluster is having samples that are different from those initially chosen at random. From the new clustering of samples, new centroids are again generated. The process of generating new clusters, and then, new centroids, is repeated until the new clusters that are generated are completely the same as in the preceding iteration \u0026ndash; this final stage of the iterative process is called convergence. The number of iterations leading to convergence of the k-means algorithm can be recorded.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003eRemoval of Potential Outliers using k-means\u003c/h2\u003e\u003cp\u003eA small experiment is run to show the importance of using a MDS plot to remove samples that evidently seem distant from the rest of samples of their phenotype, or samples that clearly mingle within the contrast phenotype. It is shown that the described samples can significantly shift the centroid of one cluster, leading to several samples not being categorized in their parent group.\u003c/p\u003e\u003cp\u003eIn order to see the effect of samples that were clearly seen as outliers in MDS plots on k-means clustering, k-means is performed in presence of those samples apart from samples s326_M_Fr and s525_F_Yo that were outstandingly distant. The k-means clustering that involves generating two initial centroids from a random selection of the remaining 48 samples consistently shows that all phenotypically young samples belong to the same cluster, except s514_F_Yo that always belongs to the same cluster as frail samples. Another noticeable outcome is that while most phenotypically frail samples also belong to the same cluster, 9 samples, which are s323_M_Fr, s315_M_Fr, s317_M_Fr, s318_F_Fr, s329_F_Fr, s302_M_Fr, s303_M_Fr, s304_M_Fr, and s306_M_Fr consistently show a belonging to the cluster having young samples. The sample s514_F_Yo which is the only phenotypically young sample that is clustered among samples that are phenotypically young lies significantly distant from the rest of frail samples in the MDS plot. This sample may be influencing the centroid of young samples by shifting it towards the frail samples, explaining why there is a consistent number of frail samples in the young group. The number of iterations in the clustering is about 5 to 10. The noted consistency is true within 12 consecutive runs of the k-means clustering.\u003c/p\u003e\u003cp\u003eTaking off s514_F_Yo leads to altered results in the clustering. After removing s514_F_Yo from the dataset, and conducting the clustering iterative process in 12 independent runs, all phenotypically young samples belong to one and the same cluster. Frail samples that consistently intervene with the cluster of phenotypically young samples during all the 12 k-means runs are now fewer (5): s323_M_Fr, s318_F_Fr, s302_M_Fr, s303_M_Fr, and s306_M_Fr. And samples s315_M_Fr and s317_M_Fr intervene with the cluster of young samples only in 1 run out of the 12 runs. Among these 5 samples, s323_M_Fr was determined to be an outlier from observations of the MDS plots; samples s318_F_Fr and s303_M_Fr seem to be on the borderlines between frail and young samples of the MDS plot in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The noticed alteration in results shows how an outlying sample can influence the clustering of samples in k-means.\u003c/p\u003e\u003cp\u003eRemoving the frail samples identified above in k-means as intervening, and also removing s514_F_Yo, leaves 18 frail samples and 24 young samples. This surviving dataset is again submitted to k-means to identify samples s508_M_Yo and s530_F_Yo as intervening with the cluster of frail samples: in 9 out of 12 runs for s508_M_Yo, and in 7 out of 12 runs for s530_F_Yo. All the remaining samples cluster within their phenotypic group in all the 12 runs. Samples s508_M_Yo and s530_F_Yo were determined to be outliers from visualization of the MDS plot in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. At this stage, all the samples that were determined as outliers from visualization of MDS plots have also been determined to be intervening in k-means.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003eSelection of Relevant Genes in ReliefF\u003c/h2\u003e\u003cp\u003eReliefF is the best known and most utilized Relief-based algorithm that uses \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e nearest hits and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e nearest misses in the weight update for each instance of the training samples, weights the instances equally, and takes the average over the number of instances used [5,10]. Relief-based algorithms are independent of the induction algorithm, i.e. after selection of a feature subset, any suitable classification scheme can be employed. This relieves of the need for the computational intensity and complexity that is involved in having to iteratively generate a feature subset until the classification seems satisfactory as in wrapper methods and embedded methods that are induction algorithm dependent. For example, there are wrappers for Na\u0026iuml;ve Bayes and Support Vector Machine classification algorithms; and in embedded methods such as decision-tree based algorithms, feature selection is part of the classification algorithm.\u003c/p\u003e\u003cp\u003eOnly genes above a threshold weight \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W\\)\u003c/span\u003e\u003c/span\u003e are selected. A random number of samples chosen from training samples is used to perform selection, and for each sample, there is an equal fixed number of nearest samples hits and misses to be considered. In the ReliefF algorithm, genes are deemed relevant to providing a contrast between two phenotypic conditions by exercising a weight that is above the threshold weight \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\delta\\:W\\)\u003c/span\u003e\u003c/span\u003e. A randomly chosen number of samples is used to determine the weights for each gene. For each gene, the weight is assessed as a cumulative sum over the randomly chosen samples. For each chosen sample, the weight is calculated for every nearest-neighbor sample hit or miss of the sample, and is then summed over the nearest neighbors. In the summation, the weight is assigned a negative value for every hit, and a positive value for every miss. A sample hit is a sample of the same phenotype as the chosen sample, while a sample miss is one of contrasting phenotype. The gene weight between the chosen sample and any other sample is calculated as the absolute value of the difference between the gene counts, divided by a constant value for that gene. That constant value is the difference between the highest count and the smallest count for the gene over all the training samples. The number of nearest hits and misses used to determine the weights is generally set here at 7 or above for the large number of training samples per phenotype. Initially, the weight for each gene is zero (0), as it accumulates as a sum over the randomly chosen samples, and for each sample, it accumulates over the nearest neighbor hits and misses. The sum is then averaged over the product of the number of randomly chosen samples and the number of nearest hits or misses. This sum can be summarized as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{W}_{n}\\)\u003c/span\u003e\u003c/span\u003e in \u003cb\u003eEq.\u0026nbsp;(2)\u003c/b\u003e. It is only different from the sum depicted in Ref. [5] in the sense that there is only one class of contrasting phenotypes.\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:{W}_{n}=\\:{W}_{n}-\\:\\sum\\:_{j=1}^{k}\\frac{\\text{d}\\text{i}\\text{f}\\text{f}\\left(n,{X}_{i},{H}_{j}\\right)}{k}\\:+\\sum\\:_{j=1}^{k}\\frac{\\text{d}\\text{i}\\text{f}\\text{f}\\left(n,{X}_{i},{M}_{j}\\right)}{k}\\:\\:\\:\\text{f}\\text{o}\\text{r}\\:i\\:\\in\\:\\left[1,m\\right]\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(2\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eIn \u003cb\u003eEq.\u0026nbsp;(2)\u003c/b\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e is the index for a gene, and runs from 1 to \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:N\\)\u003c/span\u003e\u003c/span\u003e, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:N\\)\u003c/span\u003e\u003c/span\u003e is the number of genes. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:i\\)\u003c/span\u003e\u003c/span\u003e is the index for the random chosen samples, and runs from 1 to \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:m,\\)\u003c/span\u003e\u003c/span\u003e the number of the randomly chosen samples. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{X}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the count for gene \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e in the random sample of index \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:i\\)\u003c/span\u003e\u003c/span\u003e. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:j\\)\u003c/span\u003e\u003c/span\u003e is the index for the nearest sample hit or miss, and ranges from 1 to \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e \u0026ndash; the number of nearest hits or misses. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{H}_{j}\\)\u003c/span\u003e\u003c/span\u003e is the count for gene \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e in the sample hit of index \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:j\\)\u003c/span\u003e\u003c/span\u003e, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{M}_{j}\\)\u003c/span\u003e\u003c/span\u003e is the count for gene \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e in the sample miss of index \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:j\\)\u003c/span\u003e\u003c/span\u003e. The function \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{d}\\text{i}\\text{f}\\text{f}\\left(n,X,Y\\right)\\)\u003c/span\u003e\u003c/span\u003e for any two samples is defined in \u003cb\u003eEq.\u0026nbsp;(3)\u003c/b\u003e. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{max}\\left(n\\right)\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{m}\\text{i}\\text{n}\\left(n\\right)\\)\u003c/span\u003e\u003c/span\u003e are the maximum and minimum values of the counts of gene \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e across the training samples; as such, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{d}\\text{i}\\text{f}\\text{f}\\left(n,X,Y\\right)\\:\\le\\:1\\)\u003c/span\u003e\u003c/span\u003e.\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:\\text{d}\\text{i}\\text{f}\\text{f}\\left(n,X,Y\\right)=\\:\\frac{\\left|X\\left(n\\right)-Y\\left(n\\right)\\right|}{\\text{max}\\left(n\\right)-\\text{m}\\text{i}\\text{n}\\left(n\\right)}\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(3\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eNote that the division by \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e in \u003cb\u003eEq.\u0026nbsp;(2)\u003c/b\u003e occurs \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:m\\times\\:k\\)\u003c/span\u003e\u003c/span\u003e times, and, thus setting a limit to the value of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{W}_{n}\\)\u003c/span\u003e\u003c/span\u003e: \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{W}_{n}\\:\\le\\:1\\)\u003c/span\u003e\u003c/span\u003e. The random samples are chosen by first generating a random number that is of equal distribution between 0 and 1. The generated number is then linearly projected to a number that is in the range [1, \u003cem\u003em\u003c/em\u003e], where \u003cem\u003em\u003c/em\u003e is the number of random samples to use.\u003c/p\u003e\u003cp\u003eTo determine the nearest hits and misses for a sample, the Euclidean distance is determined between the sample and the remainder of the samples. The Euclidean distance accounts for all genes in the dataset as independent features. The samples of either phenotype with the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e shortest Euclidean distances from the random sample of interest become the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e nearest sample hits and sample misses. The Euclidean distances between samples are computed a priori and stored in a table. The table is a matrix of elements which are the Euclidean distances between samples that are the corresponding row and column names of the matrix. The distance between a sample and itself is zero (0), as this makes the diagonal elements of the matrix all equal to zero.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003eGene Correlation\u003c/h2\u003e\u003cp\u003eBefore implementing the R code that determines gene correlation, the relevant genes are ranked in order of relevance. A null vector that will contain the classifier genes is initially defined in the R code. It is null, because it has no elements initially, but will increase in size as non-correlating genes are found and appended to it. The R code is written and run to produce the classifier genes based on the condition that all genes that correlate with the topmost relevant gene based on the criterion \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:R\u0026gt;\\:{R}_{th}\\)\u003c/span\u003e\u003c/span\u003e are eliminated from the set of relevant genes in the first (1st ) iteration. After the 1st iteration, the topmost relevant gene is appended to the null vector as its first element, and the gene is also discarded from the running set of relevant genes. This leaves the correlating gene that is closest by relevance to the topmost relevant gene as the top in the list of running relevant genes. The iterative process is repeated until there is nothing left in the running set of relevant genes, as all the non-correlating genes have migrated to the vector of classifier genes.\u003c/p\u003e\u003cp\u003eBesides producing the classifier genes, the top 5 genes that are closest in relevance to each classifier gene are produced. This is possible because the indices of the classifier genes in the vector of the original relevant genes are stored during the iterative correlation process. These indices are references to producing the mentioned top 5 genes. The top 5 supplementary genes are potential replacements for the specific classifier gene in case it is absent from the test dataset. A Table of the classifier genes alongside their top 5 relevant correlating genes is produced for each of the five different cases A-E leading to the obtention of classifier genes; see Table\u0026nbsp;\u003cspan refid=\"Tab11\" class=\"InternalRef\"\u003e6\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;1\u003c/b\u003e. \u003cb\u003eThe classifier genes in the same row with their top 5 relevant correlating genes in case A\u003c/b\u003e\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eclassifier gene\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGene_1\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGene_2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGene_3\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eGene_4\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGene_5\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSEMA6A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eUNC13C\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTEX56P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eC10orf71-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePCDH9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGTF2IP20\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCDC26P1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFAM153A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eHMGN2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePCSK7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCAMKMT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eENTREP3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRADD-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFAM98A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCALM2P2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePCLO\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eUCHL5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSLC26A3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVWC2L\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBACE1-AS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCERS5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSUB1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eHNRNPKP2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLINC03040\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLRRC3B-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGRAP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eERAP1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eC13orf46\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eP4HB\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eAPOBEC3C\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCYP3A43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOAS3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eC6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTMEM86B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eINPP5F\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eABCB7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMIR5697\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eACAD8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eIRF1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLARS2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eGIMAP8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eFABP5\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFAM178B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLTA4H\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eEHMT2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eDHRS4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eIMPACT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eHTRA3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCFAP107\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSETD1B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eISOC2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLIMD1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePDPR2P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eCRIM1-DT\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRPS2P32\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLAP3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eZNF407\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePAXIP1-DT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eALDH9A1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSPIN4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNACA2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIL6R-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNDUFB6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eAGFG2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNUP188\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eC1orf127\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHMGB1P37\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNUDT12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRANBP6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLSR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCYP3A5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTBC1D5\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab8\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e\u0026thinsp;\u0026minus;\u0026thinsp;2. The classifier genes in the same row with their top 5 relevant correlating genes in case B\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eclassifier gene\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGene_1\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGene_2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGene_3\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eGene_4\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGene_5\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eC10orf71-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTEX56P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eUNC13C\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGSCAR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eITGA1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGTF2IP20\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRANBP17\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRNF135\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMT-CO2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFAAH\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eDMAP1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eMORC2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRADD-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eVN1R81P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSPTBN1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGLOD4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTRIOBP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSNHG9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCDC26P1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGSTZ1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRPRD2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eDYNLT1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSEMA3E\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eERFE\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRTL9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eINPP5F\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKIF6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGAPDHP70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCOL5A1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eMRPL33\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCYP3A43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePET117\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eEFCAB14\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFAM98A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eGOLGA6GP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eZBTB44\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMIR3682\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTET1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eUQCC2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eWFS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMORN2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eNMB\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRPS2P32\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIGDCC4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCCNB3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eCDIP1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eLIMD1-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eNAA35\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCFAP107\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePLCE1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRN7SL269P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eENPP1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNBPF1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSTEAP2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRNA5SP466\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eANKLE1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAP1G2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBTG1-DT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCNR1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eEIF2B5-DT\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab9\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;3\u003c/b\u003e. \u003cb\u003eThe classifier genes in the same row with their top 5 relevant correlating genes in case C\u003c/b\u003e\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eClassifier gene\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGene_1\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGene_2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGene_3\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eGene_4\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGene_5\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eC10orf71-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSTRADB\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSKAP2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eOIP5-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eUNC13C\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eITGA1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRADD-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCLNS1A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDRD4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eATG13\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTM2D3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003ePSMD14\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCDC26P1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSMIM10L1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eUSP15\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLGR6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMRPL33\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eCCNY\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGSG1L\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLINC02937\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSNORA33\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eADCY10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eRANBP10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eCALML6\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCYP3A43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNDUFAF1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTRNAU1AP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eCOX18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eANKMY1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eKIF9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLRRC3B-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePOLR2C\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eARVCF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eCTHRC1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eEPB41L4B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eCENPV\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRPS2P32\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eKALRN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCNTF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePYROXD2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTNPO1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eRBX1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCXCL3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCENPW\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSF3B6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eILRUN-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMESDP1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSORT1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFAM238B\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNBDY\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFGGY-DT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNDUFB10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eOR7E53P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTMEM79\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCFAP107\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHOMER3-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePLEKHN1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eC16orf86\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eHSPA8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003ePABPN1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab10\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;4\u003c/b\u003e. \u003cb\u003eThe classifier genes in the same row with their top 5 relevant correlating genes in case D\u003c/b\u003e\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eClassifier gene\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGene_1\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGene_2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGene_3\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eGene_4\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGene_5\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eC10orf71-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eUNC13C\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSKAP2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTEX56P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eITGA1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eRPS27L\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRADD-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCBR1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eACTR5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePPIC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTARS2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003ePPFIBP1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCYP3A43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eKIAA2013\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eHMGXB4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eKCNE2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCCNF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eKRT7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCDC26P1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCD27-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eZNF580\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eMAP2K1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eRPRD1A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eZNF362\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGSG1L\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTRIM52-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKCNQ5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePHC1P1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTHUMPD2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003ePPIH\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRANBP17\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTSLP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePDZD11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eALDH4A1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCD38\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLINC03040\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRTL9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFLCN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMMP24OS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNUDT11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNOTCH4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTPM2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRPS2P32\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePCDHGA1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLIPT2-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLSM12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eRUBCNL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eCTDNEP1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCFAP107\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMAN1A2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eZNF888\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNLGN4Y\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePBX3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTAS2R20\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab11\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e\u0026thinsp;\u003cb\u003e\u0026minus;\u0026thinsp;5\u003c/b\u003e. \u003cb\u003eThe classifier genes in the same row with their top 5 relevant correlating genes in case E\u003c/b\u003e\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eClassifier gene\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGene_1\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGene_2\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGene_3\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eGene_4\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGene_5\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eC10orf71-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOIP5-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eUNC13C\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eITGA1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTEX56P\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSTRADB\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCDC26P1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eC1orf21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAPBB3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePDP1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eNOC2L\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eSLC44A2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRTL9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMAP4K4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGUCY1A2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eATF7IP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMCM7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eCAST\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCRADD-AS1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSCOC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBOD1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eMAPKAPK2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSEPTIN5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLINC01311\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLINC01121\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eERG\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eETNPPL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSPTBN4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eATP5IF1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGADD45G\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCYP3A43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCPNE1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAAMP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRBM39\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eUBL3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eNAB2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRANBP17\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRENBP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSLC48A1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSYNC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVEZT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eGSPT1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRPS2P32\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eATP5ME\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePIGC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eAPOOL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ePMS2CL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eXK\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSHISA8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRAD23A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDBT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSYNE4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eBAALC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eFDX1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRNA5SP466\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCFAP20\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLYSMD3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ePARP2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eRFPL3S\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eZKSCAN3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003eProcedures of kNN Sample Classification\u003c/h2\u003e\u003cp\u003eA validation dataset that is composed of the uncorrelated genes only for the validation sample is produced. Each validation sample is tested for frailty by calculating the Euclidean distance between the validation sample and an equal number of training frail and young samples that are randomly chosen. For each validation sample, a table of training samples and their Euclidean distances from the validation sample is produced. Also produced is the average of the distances of frail samples from the validation sample, and the average of the distances of young samples from the validation sample. In one classification scheme, the sample classifies as the phenotype of the nearest sample to it by Euclidean distance. In another classification scheme, the sample classifies by the phenotype of samples that are on average closer by Euclidean distance to it.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec21\" class=\"Section2\"\u003e\u003ch2\u003eTest Dataset\u003c/h2\u003e\u003cp\u003eThe test dataset is composed of normalized and annotated counts of RNA-seq data obtained from PBMCs from middle-aged non-frail (mean age\u0026thinsp;=\u0026thinsp;48) and frail (mean age\u0026thinsp;=\u0026thinsp;47.8) African Americans and whites. The dataset was deposited in the Gene Expression Omnibus (GEO) database under the title \u0026ldquo;Frailty in middle age is associated with race-specific changes to the transcriptome\u0026rdquo;, made public in August 21, 2019, and accessible from the Gene Expression Omnibus (GEO) with accession number GSE129534. The total number of samples in the dataset is 16, with each sample coming from a separate individual. The individuals are phenotypically categorized as frail or non-frail.\u003c/p\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eThe entire work featuring in the manuscript has been authored by a single individual, Marshall Tabetah, initialized as MT.Conceptualization: M.T.; Methodology: M.T.; Data Processing \u0026amp; Analysis: M.T.; Investigation: M.T.; Original Draft: M.T.; Review \u0026amp; Editing: M.T.; Figures and Visualization: M.T.; Supervision: M.T.; Funding: The authors received no funding for this work\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eMarshall Tabetah acknowledges Genelab AWG Frailty Group for useful discussions that motivated a search for frailty classifier genes.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe dataset that is used to construct the set of classifier genes is raw counts of RNA-seq data obtained from muscle biopsies of vastus lateralis, deposited in the Gene Expression Omnibus (GEO) database under the title \u0026ldquo;Muscle function in fit and (pre-)frail males and females\u0026rdquo;, made public in Feb 04, 2023, and accessible from the Gene Expression Omnibus (GEO) with accession number [GSE144304](https:/www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE144304) . The RNA-seq counts were obtained with the Illumina NextSeq 500 (Homo sapiens) platform from samples that were submitted in Jan 27, 2020. The total number of samples in the dataset is 80, with each sample coming from a separate individual. The individuals are phenotypically categorized as young, fit, or frail. Old subjects ( \u0026amp;gt; 75 years) were screened for frailty using the Fried criteria. (e.g. weight loss, exhaustion, low physical activity, slowness and weakness) [].\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003e D.B. Dongare, S.S. Nishad, S.Y. Mastoli, S.A. Saraf, N. Srivastava, A. Dey, High throughput sequencing: a breakthrough in molecular diagnosis for precision medicine, Functional \u0026amp; Integrative Genomics (2025) 25:22 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10142-025-01529-w\u003c/span\u003e\u003cspan address=\"10.1007/s10142-025-01529-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e J.M. Churko, G.L. Mantalas, M.P. Snyder, and J.C. Wu, Overview of High Throughput Sequencing Technologies to Elucidate Molecular Pathways in Cardiovascular Diseases, Circ Res. 2013 June 7; 112(12): doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIRCRESAHA.113.300939\u003c/span\u003e\u003cspan address=\"10.1161/CIRCRESAHA.113.300939\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e J-Y Lee, The Principles and Applications of High-Throughput Sequencing Technologies, \u003cem\u003eDev Reprod\u003c/em\u003e 2023;27(1):9\u0026ndash;24 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.12717/DR.2023.27.1.9\u003c/span\u003e\u003cspan address=\"10.12717/DR.2023.27.1.9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e \u003cem\u003eAmer\u003c/em\u003e F.A.H. ALNUAIMI and \u003cem\u003eTasmin\u003c/em\u003e H.K. ALBALDAWI, An overview of machine learning classification techniques, BIO Web of Conferences \u003cb\u003e97\u003c/b\u003e, 00133 (2024) \u003cem\u003eISCKU 2024\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1051/bioconf/20249700133\u003c/span\u003e\u003cspan address=\"10.1051/bioconf/20249700133\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e \u003c/li\u003e\u003cli\u003e\u003cspan\u003e Z. Yaping, Z. Changyin, Gene Feature Selection Method Based on ReliefF and Pearson Correlation, 2021 3rd International Conference on Applied Machine Learning (ICAML). DOl 1O.1109/1CAML54311.2021.00011\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e X. Cao, L. Ding and T.B. Mersha, Development and validation of an RNA‑seq‑based transcriptomic risk score for asthma, Scientific Reports (2022) 12:8643 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-022-12199-0\u003c/span\u003e\u003cspan address=\"10.1038/s41598-022-12199-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Bostanci, E.; Kocak, E.; Unal, M.; Guzel, M.S.; Acici, K.; Asuroglu, T. Machine Learning Analysis of RNA-seq Data for Diagnostic and Prognostic Prediction of Colon Cancer. Sensors \u003cb\u003e2023\u003c/b\u003e, 23, 3080. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s23063080\u003c/span\u003e\u003cspan address=\"10.3390/s23063080\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Z. Yaping, Z. Changyin, Gene Feature Selection Method Based on ReliefF and Pearson Correlation, 2021 3rd International Conference on Applied Machine Learning (ICAML). DOI1O.1109/1CAML54311.2021.00011\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e J. Perez-Ortega, N.N. Almanza-Ortega, A. Vega-Villalobos, R. Pazos-Rangel, C. Zavala-Diaz, and A. Martinez-Rebollar, The K-Means Algorithm Evolution, \u003cem\u003eDOI\u003c/em\u003e: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://dx.doi.org/10.5772/intechopen.85447\u003c/span\u003e\u003cspan address=\"10.5772/intechopen.85447\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e R.J. Urbanowicz, M. Meeker, W. La Cava, R.S. Olson, J.H. Moore, Relief-based feature selection: Introduction and review, \u003cem\u003eJournal of Biomedical Informatics 85 (2018) 189\u0026ndash;203\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jbi.2018.07.014\u003c/span\u003e\u003cspan address=\"10.1016/j.jbi.2018.07.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e M. I. Love, W. Huber, S. Anders: Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. \u003cem\u003eGenome Biology\u003c/em\u003e 2014, \u003cb\u003e15\u003c/b\u003e:550. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://dx.doi.org/10.1186/s13059-014-0550-8\u003c/span\u003e\u003cspan address=\"10.1186/s13059-014-0550-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e G.J. Szekely, M.L. Rizzo, and N.K. Bakirov, Measuring and Testing Dependence by Correlation of Distances, The Annals of Statistics 2007, Vol. 35, No. 6, 2769\u0026ndash;2794. DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1214/009053607000000505\u003c/span\u003e\u003cspan address=\"10.1214/009053607000000505\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e J. Sun, J.D. Herazo-Maya, X. Huang, N. Kaminski, H. Zhao, Distance correlation based gene set analysis in longitudinal studies, Statistical Applications in Genetics and Molecular Biology. 2018; 20170053.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Hou \u003cem\u003eet al.\u003c/em\u003e Distance correlation application to gene co-expression network analysis, \u003cem\u003eBMC Bioinformatics\u003c/em\u003e (2022) \u003cem\u003e23:81\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12859-022-04609-x\u003c/span\u003e\u003cspan address=\"10.1186/s12859-022-04609-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e R.A. Tumasian III, A. Harish, G. Kundu, J-H Yang, C. Ubaida-Mohen, M. Gonzalez-Freire, M. Kaileh, L.M. Zukley, C.W. Chia, A. Lyashkow, W.H. Wood III, Y. Piao, C. Coletta, J. Ding, M. Gorospe, R. Sen, S. De \u0026amp; L. Ferruci, Skeletal muscle transcriptome in healthy aging, NATURE COMMUNICATIONS (2021) 12:2014 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41467-021-22168-2\u003c/span\u003e\u003cspan address=\"10.1038/s41467-021-22168-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e \u003c/li\u003e\u003cli\u003e\u003cspan\u003e Fried, L.P., Tangen, C.M., Walston, J., et al. (2001) Cardiovascular Health Study Collaborative Research Group. Frailty in Older Adults: Evidence for a Phenotype. The Journals of Gerontology Series A: Biological Sciences and Medical Sciences, 56, M146-M156. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://dx.doi.org/10.1093/gerona/56.3.M146\u003c/span\u003e\u003cspan address=\"10.1093/gerona/56.3.M146\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e M.I. Love, S. Anders, W. Huber: Differential analysis of count data \u0026ndash; the DESeq2 package (2016)\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"frailty, phenotype, genotype, classifier genes, relevant genes, correlation, training samples, validation samples, machine learning, k-means, test samples, misfits, clustering, classification","lastPublishedDoi":"10.21203/rs.3.rs-7475052/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7475052/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eA set of frailty classifier genes is constructed based on a RNA-seq dataset with phenotypically contrasting frail and non-frail samples. Authentication of the obtained classifier genes as a generic set that will be applicable to new test samples is verified by testing whether frail and non-frail samples are separately identified in a test dataset. Verification of the gene selection approach is performed by using both a supervised learning approach, k-nearest neighbor (kNN), and an unsupervised learning approach, k-means, to classify test samples, as the different approaches show remarkable coincidence in some results. The numbers of classifier genes that closely reproduce the classification outcomes based on entire gene sets are remarkably small, only\u0026thinsp;~\u0026thinsp;10 genes. The classifier genes are shown to correlate with several notable genes in the literature. An online search of gene sets that overlap with the top 500 relevant genes to frailty reveals gene ontologies that associate with frailty.\u003c/p\u003e","manuscriptTitle":"RNA-Seq Data Analyses and Machine Learning Predict Frailty Classifier Genes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-27 11:12:20","doi":"10.21203/rs.3.rs-7475052/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"6f57f224-93e3-452b-a7d8-08b2d04f78d9","owner":[],"postedDate":"October 27th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":56807713,"name":"Health sciences/Biomarkers"},{"id":56807714,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":56807715,"name":"Biological sciences/Genetics"}],"tags":[],"updatedAt":"2025-10-27T11:12:21+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-27 11:12:20","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7475052","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7475052","identity":"rs-7475052","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}