Machine Learning Integration of Tissue-Specific Metagenomic Signatures for Colorectal Cancer Diagnosis

Machine Learning Integration of Tissue-Specific Metagenomic Signatures for Colorectal Cancer Diagnosis | 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 Research Article Machine Learning Integration of Tissue-Specific Metagenomic Signatures for Colorectal Cancer Diagnosis Anıl DELİK, Yakup Ülger, Ferhat Albayrak, Umut Orhan, Ulku Unal, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7808541/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Background Colorectal cancer (CRC) represents a significant global health burden. Leveraging machine learning (ML) with metagenomic and tissue-specific data presents new opportunities for improving diagnostic accuracy and understanding the microbiome's role in CRC. Objective This study aimed to enhance diagnostic efficiency and identify crucial bacterial biomarkers in CRC using various ML models applied to metagenomic data. Material and Methods A total of 33 samples were analyzed, comprising 20 healthy controls and 13 CRC patients. Each sample included demographic data (age, gender) and bacterial information ( Bacteriodes , Enterococcus , Faecalibacterium , Proteobacteria , Gammaproteobacteria , Firmicutes , Enterobacteriaceae , Clostridia ). We employed five ML models: Linear Regression, Naive Bayes, Multilayer Perceptron (MLP), Support Vector Machine (SVM), and Decision Tree. Using leave-one-out cross-validation, we evaluated model accuracy and optimized feature selection by iteratively removing bacterial data to improve model performance. Results The highest accuracy was 87.88% in the presence of demographic information, while the highest accuracy was 84.85% in the absence of demographic information. In the presence of demographic information, the highest accuracy was achieved with the MLP model, and in the absence of demographic information, with the SVM linear model. While the MLP model required 5 bacterial information to reach a accuracy of 87.88%, the SVM linear model could reach a accuracy of 84.85% with 6 bacterial information. Conclusions Machine learning models, particularly MLP, combined with metagenomic data, can significantly improve CRC diagnostic accuracy. Identifying key bacterial biomarkers enhances our understanding of the microbiome's role in CRC, providing a foundation for non-invasive diagnostic tools and personalized treatment strategies. colorectal cancer metagenomic machine learning diagnosis prognosis Figures Figure 1 Figure 2 Figure 3 Introduction Colorectal cancer (CRC) is a major public health concern worldwide, characterized by high morbidity and mortality rates (Arnold et al., 2017 ; Siegel et al., 2020 ). Early detection and accurate diagnosis are pivotal for effective treatment and improved survival rates. Traditional diagnostic methods, though effective, often face limitations in sensitivity and specificity (Dekker et al., 2019 ). Therefore, there is a pressing need for innovative approaches that can enhance diagnostic precision and provide a deeper understanding of the disease's underlying mechanisms. Recent advancements have highlighted the significant role of the gut microbiome in the pathogenesis and progression of CRC (Wang et al., 2023 ). The gut microbiota, composed of trillions of microorganisms, plays a crucial role in maintaining intestinal homeostasis, modulating immune responses, and influencing inflammatory processes (Rooks and Garret, 2016; Kho and Lal, 2018 ). Dysbiosis, or the imbalance in microbial composition, has been linked to the development and progression of various cancers, including CRC (Anderson and Sears, 2023 ; Zeng et al., 2024 ). Specific alterations in microbial communities can lead to chronic inflammation, immune dysregulation, and even direct interactions with tumor cells, promoting carcinogenesis (Rebersek et al., 2021). Leveraging machine learning (ML) to analyze metagenomic data offers a promising avenue for enhancing CRC diagnostics (Liu et al., 2024 ). ML models can process vast amounts of data, identifying patterns and biomarkers that might be overlooked by traditional statistical methods. In the context of CRC, ML can be employed to analyze both microbiome and tissue-specific data, providing a comprehensive understanding of the host-microbiome interactions and their implications for CRC (Freitas et al., 2023 ). This study aims to investigate the potential of ML-based approaches to enhance diagnostic efficiency in CRC by analyzing metagenomic and tissue-specific data. By employing various ML models, including Linear Regression, Naive Bayes, Multilayer Perceptron (MLP), Support Vector Machine (SVM), and Decision Tree, we seek to identify the most effective methods for distinguishing CRC patients from healthy controls. Additionally, we aim to pinpoint the key bacterial taxa that serve as biomarkers for CRC, thereby contributing to the development of non-invasive diagnostic tools and personalized treatment strategies. We hypothesize that ML models, particularly those incorporating demographic and bacterial information, will significantly improve CRC diagnostic accuracy. The iterative feature selection process, aimed at optimizing model performance by excluding less informative bacterial data, is expected to highlight the most critical bacterial biomarkers. This approach not only enhances our understanding of the microbiome's role in CRC but also provides a robust foundation for future clinical applications. Materials and Methods Sample Collection This study includes tissue samples from patients diagnosed with CRC and healthy individuals who visited the Gastroenterology Clinic at Çukurova University Balcalı Hospital Faculty of Medicine. The samples from both patients and the control group were collected in accordance with ethical committee approval and informed consent obtained from the participants. Ethics and Consent Approvals This research received approval from the Ethics Committee of the Çukurova University Faculty of Medicine on November 3, 2023 (IRB numbers: 2023-26-138). Written consent was obtained from all participants. The study was conducted in alignment with the Declaration of Helsinki and adhered to relevant ethical guidelines. Patient Selection CRC patients included in the study were individuals with a histopathologically confirmed diagnosis of CRC, ranging from stage I to IV. Patients were compared with a matched healthy control group based on parameters such as age, gender, and cancer stage. Tissue samples from both groups were collected during colonoscopy procedures. Exclusion Criteria Previously Diagnosed Malignant Diseases: Patients who have been previously diagnosed or treated for another malignant disease. Severe Comorbid Conditions: Individuals with severe comorbid conditions such as uncontrolled hypertension, heart failure, or active infection. Medication Use: Patients who have taken immunosuppressive drugs or chemotherapy within the last three months. Factors Affecting Microbiota: Individuals who have used antibiotics, probiotics, or prebiotic supplements within the last six months. Insufficient Sample Quality: Samples that were insufficient or contaminated for DNA isolation or sequencing. Lack of Ethical Approval: Individuals without ethical committee approval or who do not adhere to study protocols. Lack of Informed Consent: Individuals who did not sign the informed consent form. Non-Compliance with Study Protocol: Individuals with demographic characteristics such as age, gender, or stage that do not comply with the study protocol. Metagenomic Data Analysis DNA was isolated from the collected tissue samples and prepared for high-resolution metagenomic sequencing. The obtained sequencing data were quality controlled and subsequently used for the identification and genetic profiling of microbial species. 16S rRNA gene sequencing The V3-V4 region of the 16S rRNA gene was amplified with 341F (5’-CCTACGGGNGGCWGCAG-3’) and 805R (5’-GACTACHVGGGTATCTAATCC-3’) primers. Sequence results were obtained using the Illumina MiSeq™ platform. Sequence data were automatically converted to FASTQ format by the MiSeq™ instrument. Data analysis was performed using QIIME2 software. Greengenes2 database was used for identification and taxonomic assignment of microbiota sequences. Relative abundances of microbiome OTUs Relative abundance plots were generated using the R programming language to show how the specimens varied at which taxonomic level. These plots show the diversity of the samples in terms of percentages, showing the variation of the prominent organisms. The 'tidyverse' library (Wickham et al., 2019 ), which provides a large collection of packages for data manipulation and visualization, and the 'RColorBrewer' library (Neuwirth, 2014 ) which provides color palette options in R, were used to visualize data consisting of the taxonomic levels of phylum, class, family and genus. Normalization and Preprocessing of Metagenomic Data Metagenomic data often contain a high proportion of zero values, so methods were applied to minimize the zero inflation problem. Log Transformation (Log10 or Natural Log): To correct extreme skewness, very small values were adjusted by adding + 1 before log transformation. CLR (Centered Log-Ratio) Transformation: Given the compositional nature of metagenomic data, they are not treated as independent variables, so a ratio-based transformation was used. TMM (Trimmed Mean of M-values) Normalization: Applied to generalize abundance data and reduce sample-based differences. RA (Relative Abundance) Transformation: Bacterial abundance was normalized against total abundance to maintain comparability. VST (Variance Stabilizing Transformation): Used to allow an unbiased evaluation of biological differences between CRC and healthy groups. Negative Binomial Model: Tools like EdgeR and DESeq2 were employed to analyze bacterial abundance and balance overdispersion. Z-score Standardization: Applied to adapt bacterial abundances to a standardized scale for better comparability. After transformation, a feature selection process was conducted to identify the most meaningful biomarkers. Recursive Feature Elimination (RFE) and SHAP (Shapley Decomposition) were integrated into the modeling process to provide explainable AI-based feature selection. Alpha diversity in metagenome analysis Alpha diversity is a metric that used to assess the diversity of ecological communities and refers to the richness of species found in a given habitat or sampling site (Oh et al., 2021 ). In the alpha diversity analysis, the Shannon indexes ( H’ ) (Spellerberg & Fedor, 2003 ) were calculated at the genus taxonomic level as a method so that both the richness (i.e. the number of genus) and abundance (i.e. the relative distribution of the number of individuals of genus) of the genus in the samples examined were taken into account. The equation for calculating the Shannon index ( H’ ) is expressed as follows: p i is the proportion of individuals belonging to genus i in the dataset. After the calculation for each sample, the Shannon entrophy index plot was visualized in Python programming language employing 'pandas' (McKinney & Team, 2015 ) library and 'pyplot' module of 'matplotlib' library (Hunter, 2007 ). Beta diversity in metagenome analysis Beta-diversity analysis, in which similarities between microbiome pairs are quantified in terms of distance (Delik et al., 2022 ), was performed using the Bray-Curtis dissimilarity measure (Bray & Curtis, 1957 ). The relative abundance values of members of the Genus taxonomic level were used to create the distance matrix in the Python programming language. The distance matrix was obtained using the ‘pandas’(McKinney & Team, 2015 ) and ‘numpy’ libraries (Harris et al., 2020 ), as well as the 'pdist' function which calculates the distances between pairs and 'squareform' which converts the compressed distance vector into a square distance matrix in the ‘scipy.spatial.distance’ library (Virtanen et al., 2020 ). To perform principal coordinate analysis (PCoA) using the distance matrix, the MDS (Multidimensional Scaling) class in the 'sklearn.manifold' library (Pedregosa et al., 2011 ), which is used in the Python programming language to transform high-dimensional data sets into a lower dimensional space, was used and the data set was transformed with the 'fit_transform' method. The PCoA plot was visualized using the 'matplotlib.pyplot' library (Hunter, 2007 ). Machine Learning Models In the study, 5 different models and variations of these models were used: linear regression, naive bayes, multilayer perceptron (MLP), support vector machine (SVM) and decision tree. Linear Regression is a statistical method that models the relationship between a dependent variable ( \(\:y\) ) and one or more independent variables ( \(\:x\) ). The goal is to find the best linear equation that describes how the dependent variable \(\:y\) varies with respect to the independent variables \(\:{x}_{1},{x}_{2},\:{x}_{3},\dots\:,{x}_{n}\) . This equation is usually expressed as follows: $$\:y=\:{\beta\:}_{0}+{\beta\:}_{1}*{x}_{1}+{\beta\:}_{2}*{x}_{2}+\dots\:+{\beta\:}_{n}*{x}_{n}+\epsilon\:$$ 1 While the dependent variable \(\:y\) in (1) represents the estimated or in other words the desired target, the variables \(\:{x}_{1},{x}_{2},\dots\:,{x}_{n}\) represent the independent variables. Although \(\:{\beta\:}_{0}\) is a constant term, the variables \(\:{\beta\:}_{1},\:{\beta\:}_{2},\dots\:,{\beta\:}_{n}\) are the regression coefficients of the variables \(\:{x}_{1},{x}_{2},\dots\:,{x}_{n}\) and the effect of these variables on the dependent variable \(\:y\) . The variable \(\:\epsilon\:\) at the end is the error term and represents the unexplained random errors in the model. Linear regression assumes that the relationship between the independent variables and the dependent variable is linear and estimates this relationship using the ordinary least squares (OLS) method. Linear regression is widely used in fields such as data analysis, machine learning, health, and economics (Baba et al., 2010 ; Shafi & Rusiman, 2015 ). Naive Bayes is a probabilistic classification algorithm and was developed based on Bayes' theorem. Bayes' theorem is a theorem that calculates the probability of an event occurring using prior knowledge of the events that occurred before that event. Bayes' theorem is expressed as follows: $$\:P\left(A|B\right)=\:\frac{P\left(B|A\right)*P\left(A\right)}{P\left(B\right)}$$ 2 The probability \(\:P\left(A|B\right)\) in (2) shows the probability of event \(\:A\) occurring if event \(\:B\) occurs. While \(\:P\left(B|A\right)\) shows the probability of event \(\:B\) occurring if event \(\:A\) occurs, \(\:P\left(A\right)\) and \(\:P\left(B\right)\) show the probabilities of events \(\:A\) and \(\:B\) occurring. The Naive Bayes algorithm also calculates probability according to the Bayes theorem given in Eq. ( 2 ). When calculating probability, it is calculated with the assumption that all features are independent of each other. The values obtained at the end of the calculation are the probabilities that the sample belongs to the calculated class. The sample is included in the class with the highest value. The Naive Bayes method is widely used in many fields such as statistics, commerce, health, deep learning, and education (Ali et al., 2020 ; Chen et al., 2020 ; Ohata et al., 2021 ; Perez & Perez, 2021 ). In Decision Tree, each branch of tree, which works as a tree structure that makes decisions by dividing data, represents a specific decision or result. There are 3 types of nodes in decision trees: root node, internal node, and leaf nodes. The root node is the starting point of the decision tree and contains the entire dataset. Here, the feature that best separates the data is selected. Internal nodes are nodes that start from the root node and each divide the data according to a feature or situation. A test or decision is made on a feature in each internal node. Leaf nodes are the end points of the decision tree and represent a specific class or value. These nodes contain classes in classification problems and numerical values in regression problems. In addition, there are elements called branches in decision trees that represent the paths extending between nodes. Each branch represents the probability of a decision or result. Multilayer Perceptron (MLP), one of the types of artificial neural networks (ANN), has additional hidden layers in addition to the input and output layers, unlike classical ANNs. MLP is a feedforward ANN model developed to solve the XOR problem without linear separability. The MLP model consists of three-layer types: input layer, hidden layer and output layer. The number of hidden layers can be more than one, and the number of inputs \(\:{x}_{1},{x}_{2},\dots\:,{x}_{n}\) and outputs \(\:{y}_{1},{y}_{2},\dots\:,{y}_{n}\) can vary. As in non-multilayer ANNs, the input layer is the data layer that comes to the MLP cell. The incoming data is transmitted to the middle layer, also known as the hidden layer. While the increase in the number of cells in the hidden layer or layers causes the MLP to learn more accurately, it can also cause it to learn or memorize in a longer time. The output layer processes the information coming from the middle layer and produces the output of the MLP. Since its foundations in the 1960s, Support Vector Machine (SVM) is one of the best-known methods that still provides the best classification success in many applications. The algorithm, which was first developed by Vapnik and Lerner (1963), took its current state with studies at AT&T Bell Laboratory. It has become one of the standard methods in machine learning due to its extraordinary success in many different problems (Boser et al., 1992 ; Guyon et al., 1992 ; Cortes, 1995 ; Schölkopf et al., 1995; Schölkopf et al., 1996 ; Vapnik et al., 1996 ). The most important feature of SVM is support vectors. While classifying data as a SVM model, it focuses on the most critical points of the data and calls these points support vectors and uses them to define the hyperplane. The selected support vectors are the points closest to the hyperplane. Another important feature is kernel functions. Using these functions, data can be transformed into different spaces. The most used kernel functions are linear, polynomial, radial basis function (RBF) and sigmoid kernel functions. Rationale for the Selection of Eight Bacterial Features Prior to analysis, the dataset contained a total of 672 bacterial species. In the initial phase, all bacterial features were considered; however, a feature selection process was applied, taking into account high collinearity, low variance, and weak biological relevance, which could negatively impact model performance. The feature selection focused not only on classification accuracy but also on preserving taxa that would strengthen biological interpretation. Bacteria with a strongly demonstrated association with CRC in the literature were prioritized during analysis to minimize the risk of excluding key diagnostic patterns from the microbiome. Multiple biological and statistical criteria were considered in the identification of the eight selected bacteria: A literature review of published metagenomic studies and CRC biomarkers was conducted. The distribution of bacterial abundance was statistically analyzed to identify CRC-specific bacteria. Expert opinions on microbiome-CRC relationships were gathered to select features that would enhance diagnostic models. Statistical Analysis All statistical analyses were carefully designed to ensure the accurate evaluation of machine learning models and data. The methods used in the analysis of demographic and bacterial data, along with their implementation details, are presented below: Between-Group Comparisons: To examine the differences in demographic data (age, BMI) between CRC patients and the healthy control group, appropriate statistical tests were used: Age and BMI: Since the data did not meet the assumption of normal distribution, an independent two-sample t-test was not applied to compare mean differences between the two groups. Instead, the non-parametric Mann-Whitney U test was used. Gender: The chi-square test was applied to evaluate the relationship between two categorical variables (group and gender). Given the small sample size, Yates' continuity correction was performed, and its impact on the p-value was reported. Statistical Assumptions: The assumptions for tests such as the t-test and chi-square test were assessed: T-Test: Normal distribution assumptions were evaluated using the Shapiro-Wilk test, and variance homogeneity was tested using Levene’s test. Chi-Square Test: Yates' continuity correction was applied to the 2x2 contingency table, and p-values were reported with the correction applied. Summary Statistics: Age and BMI: Reported as mean ± standard deviation (SD). Gender distribution: Presented as percentages. Interpretation of P-Values: Hypothesis test p-values were interpreted based on a 5% significance level (α = 0.05). A 95% confidence interval was used in all analyses to ensure more precise comparisons between groups. Two-tailed test results were used, with no preference for one-tailed tests. Bacterial Data Analysis: Due to common metagenomic data characteristics such as skewness, zero inflation, and overdispersion, bacterial data were transformed and normalized. These transformations were performed to meet linear model assumptions and improve model performance. The comparative performance of models was evaluated before and after transformation. Machine Learning Performance Metrics: To ensure class imbalance (13 CRC vs. 20 control) did not affect model performance, additional metrics besides accuracy were used, including AUC-ROC and precision-recall scores, which provided a more comprehensive evaluation of the model’s predictive performance. Software and Tools: All statistical analyses were conducted using SPSS software. PCoA (Principal Coordinate Analysis): Bacterial data visualization was performed using Python and relevant libraries (scikit-bio, matplotlib). Results Demographic feature of study population The average age of CRC patients was 64.83 years with a standard deviation of 13.51, while the control group had an average age of 60.70 years with a standard deviation of 13.47. The difference was not statistically significant (P value = 0.20). In the CRC group, 23% were female and 77% were male. In contrast, the control group had an equal distribution of 50% for both genders. The difference in gender distribution had a P value of 0.09, indicating a trend towards significance but not statistically conclusive. The average BMI for the CRC group was 25.85 kg/m², and for the control group, it was 24.85 kg/m². The difference was not statistically significant (P value = 0.28). The majority of tumors in the CRC group were located in the rectum (53.84%), followed by the sigmoid colon (15.38%), ascending colon (15.38%), transversal colon (7.69%), and cecum (7.69%). No patients were at stage T1, 23% were at stage T2, 31% at T3, and 46% at T4, indicating a higher prevalence of advanced-stage tumors. Only 5% of the CRC cases showed invasion, while the vast majority (95%) did not (Table 1 ). Table 1 Demographic data in colorectal cancer and control individuals Variable CRC (n = 13) Control (n = 20) 95%(CI) P value Age (years),( mean, SD ) 64.83 ± 13.51 60.70 ± 13.47 (56.76–72.90)* (54.22–67.18)** 0.40 a Gender (n,%) 0.146 b Female 3 (23) 11 (55) (6.1–54.0)* (32.2–75.6)** Male 10 (77) 9 (45) (46.0–93.9)* (24.4–67.8)** BMI (kg/m2), ( mean, SD ) 25.85 ± 2.09 24.85 ± 5.51 (24.46–27.24)* (22.14–27.56)** 0.28 c Tumor location (n,%) - Rectum 7 (53.84) - Sigmoid colon 2 (15.38) - Ascending colon 2 (15.38) - Transversial colon 1 (7.69) - Cechum 1 (7.69) - Tumor staging (T) (n,%) - - T1 0 (0) - T2 3 (23) - T3 4 (31) - T4 6 (46) - İnvasion (n;%) - - Yes 1 (5) - No 19 (95) - BMI: Body mass index, SD: Standard deviation. CRC, Control. Confidence Intervals (95% CI): Calculated for age, BMI, and gender ratios and included in the table. Used to highlight variability and uncertainty in small sample sizes. Statistical Tests: a: Parametric t-test for age. c: Mann-Whitney test for BMI. b: Chi-square test for gender variable, with Yates' continuity correction applied due to small sample size. P-values: All p-values were assessed bilaterally and reported according to the appropriate tests. Tumor Location and Staging Data: As these data are not applicable for the control group, they were only reported for the CRC group. Microbiome diversity in samples According to the analyzed data, relative abundance plots were obtained for the four taxonomic levels, namely phylum, class, family, and genus (Fig. 1 ). Firmicutes , which stand out at the phylum level, had a mean value of 38.07% while Proteobacteria had a mean value of 23.11% in total samples. At the class level, Gammaproteobacteria had a mean value of 35.12% and Clostridia had a mean value of 22.49% in total samples. At the family level, Enterobacteriaceae accounted for 32.32% of the total samples. At the genus level, Bacteroides , Enterococcus , and Faecalibacterium had mean values of 12.14%, 5.14%, and 2.29%, respectively. The mean values of the microbiomes in cancer and control samples are given in Table 2 for each taxonomic level. Table 2 Proportional distribution of microbiomes in cancer and control samples Taxonomic level Taxon name Mean (Cancer) Mean (Control) Phylum Firmicutes 23.109 45.531 Proteobacteria 37.350 38.553 Class Gammaproteobacteria 21.079 44.257 Clostridia 24.237 21.368 Family Enterobacteriaceae 19.744 40.502 Genus Bacteroides 19.701 7.234 Enterococcus 10.640 1.565 Faecalibacterium 1.510 2.797 Alpha diversity of microbiomes Based on the alpha diversity analysis performed at the genus level to evaluate the diversity in the samples, cancer samples (CA13, CA10, and CA11) were determined to have the highest Shannon entropy index values, with 1.021, 0.840, and 0.820, respectively (Fig. 2 ). The lowest alpha diversity was identified in S4 and S9 with an H' index value of 0, and in S7 with an H' index value of 0.012. Mann-Whitney U test was conducted for statistical analysis, yielding a p-value of 0.077. This result indicates that the differences in entropy between the two groups are not statistically significant. Beta diversity of microbiomes Beta diversity was visualised by PcoA (Fig. 3 ). Beta diversity analysis showed that cancer and control samples could be distinguished as two distinct groups. It could be said that sample heterogeneity also had a greater effect on the overall microbiome composition in our samples. Machine learning model results with metagenomic data There are 33 samples in the study, 20 samples in healthy group and 13 samples in control group. Each sample has 10 features, 2 demographic and 8 bacterial information. Demographic information includes age and gender, bacterial information includes Bacteriodes , Enterecoccus , Faecelibacterium , Proteobacteria , Gammaproteobacteria , Firmicutes , Enterobacteriaceae and Clostridia . These features were selected according to the expert opinion according to the fullness rate, the accuracy of the data and their importance for CRC. In the study, 5 different models were used: linear regression, naive bayes, multilayer perceptron, support vector machine and decision tree. For MLP, the hidden layer, the number of neurons in the hidden layer and the optimization method were changed and tested and optimized. For SVM, tests were performed with different kernel functions. In all tests performed; Gender and class information were converted to numerical values ​​as 0 (gender for female, class for healthy group) and 1 (gender for male, class for CRC group). Due to the small number of samples, it was aimed to increase learning and prediction accuracy by using the leave-one-out cross validation method. Tests were carried out on all features, and as a result of each test, the bacteria that had a negative effect on the most models were removed and the highest accuracy was tried to be achieved with the least bacterial data. The tests were repeated with and without demographic information in the features and the effect of demographic information was examined. In the tests performed with linear regression, calculations were made by giving all the features and if the results obtained as a result of the regression were equal to or greater than 0.5, it was included in the control group, and if it was smaller, it was included in the healthy group. Studies conducted with Naive Bayes and Decision Tree methods were conducted without any pre- and post-processing. The accuracy obtained with Naive Bayes were compared with the results of other models as an anchor. It was aimed that other methods, especially deep learning methods, would show higher accuracy than the Naive Bayes method. In MLP tests, firstly, the settings that achieved the highest accuracy were selected by using the entire dataset and changing the optimization method, the number of hidden layers, the number of neurons in the hidden layer, and the maximum number of iterations. As a result of the preliminary tests, it was seen that the optimization method adam, the number of hidden layers and neurons as 1 hidden layer 100 neurons, and the maximum iteration as 500 iterations gave the highest accuracy. The settings of the MLP method were determined in this way in the study. Preliminary tests were performed on the kernel functions, which are the most important feature of the SVM method, using the entire dataset. Linear, polynomial, sigmoid and RBF functions were used as kernel functions. Since the accuracy of the sigmoid and RBF kernel functions in the obtained results remained below the desired level, it was deemed appropriate to perform tests only with linear and polynomial kernel functions in the SVM method. In the study, tests were conducted with and without demographic information using 5 models and 6 methods. As a result of these tests, it was observed that demographic information increased the accuracy in estimation. All tests were started with 8 bacteria information and by removing the bacteria information that decreased or did not increase the accuracy at each step, it was aimed to reach the highest accuracy with the least bacteria information. The first test with 8 bacteria and 2 demographic information is given in Table 3 . Table 3 Demographic information and results of Test 1 with the entire dataset. Bacteria Name Extracted from Dataset Model Accuracy (%) Linear Regression Naive Bayes Decision Tree MLP Linear Regression SVM Poly All Data 72,73 78,79 66,67 72,73 66,67 66,67 Bacteriodetes 75,76 78,79 72,73 66,67 72,73 72,73 Enterococcus 66,67 69,70 57,58 54,55 51,52 60,61 Faecalibacterium 63,64 75,76 69,70 78,79 84,85 69,70 Proteobacteria 81,82 75,76 66,67 75,76 69,70 66,67 Gammaproteobacteria 75,76 75,76 66,67 78,79 66,67 66,67 Firmicutes 63,64 78,79 63,64 69,70 66,67 60,61 Enterobacteriaceae 75,76 75,76 69,70 75,76 72,73 69,70 Clostridia 66,67 75,76 75,76 75,76 75,76 75,76 When Table 3 is examined, it is seen that the Naive Bayes method reached the highest accuracy value with 78.79% when there is demographic information and 8 bacterial information in the dataset. However, since the aim of the study was to reach the highest accuracy with the least bacteria information, each bacterium was removed from the dataset one by one, and the tests were repeated. In this case, when bacteriodetes was removed from the dataset, it was seen that the accuracy increased in linear regression, decision tree, SVM linear and SVM poly models; when Faecalibacterium or Clostridia was removed from the dataset, it was seen that the accuracy increased in decision tree, MLP, SVM linear and SVM poly models; when proteobacteria was removed from the dataset, it was seen that the accuracy increased in linear regression, MLP and SVM linear models; when Gammaproteobacteria was removed from the dataset, it was seen that the accuracy increased in linear regression and MLP models; when Enterobacteriaceae was removed from the dataset, it was seen that the accuracy increased in linear regression, decision tree, MLP, SVM linear and SVM poly models. Since the absence of Entreobacteriaceae data caused the estimation to be more accurate in the 5 models, Entreobacteriaceae was removed from the dataset and the second test was performed. The results of Test 2 are given in Table 4 . Table 4 Demographic information and results of Test 2 with 7 bacteria. Bacteria Name Extracted from Dataset Model Accuracy (%) Linear Regression Naive Bayes Linear Regression MLP Linear Regression SVM Poly All Data 75,76 75,76 69,70 75,76 72,73 69,70 Bacteriodetes 75,76 72,73 72,73 75,76 72,73 72,73 Enterococcus 72,73 66,67 63,64 69,70 66,67 66,67 Faecalibacterium 66,67 75,76 66,67 78,79 75,76 66,67 Proteobacteria 78,79 72,73 69,70 72,73 72,73 69,70 Gammaproteobacteria 75,76 72,73 69,70 75,76 69,70 69,70 Firmicutes 66,67 75,76 72,73 72,73 63,64 69,70 Clostridia 72,73 75,76 72,73 75,76 66,67 72,73 As in Test 1, there are three bacteria in Test 2 that, when removed from the dataset, increase accuracy on the most models. Among these bacteria, the removal of the bacteria named Faecalibacterium from the dataset caused an increase in accuracy in two models, while it caused a decrease in accuracy in linear regression, decision tree and SVM poly models. The removal of the bacteria named Clostridia from the dataset caused an increase in two models, while it caused a decrease in accuracy in linear regression and SVM linear models. Finally, the removal of the bacteria named Bacteriodetes caused an increase in accuracy in two models, while it caused a decrease in accuracy in only one model. For this reason, Bacteriodetes was removed from the new dataset containing 7 bacterial information and the third test was performed. The results obtained with Test 3 are given in Table 5 . Table 5 Demographic information and results of Test 3 with 6 bacteria. Bacteria Name Extracted from Dataset Model Accuracy (%) Linear Regression Naive Bayes Linear Regression MLP Linear Regression SVM Poly All Data 75,76 72,73 72,73 75,76 72,73 72,73 Enterococcus 72,73 66,67 69,70 72,73 72,73 72,73 Faecalibacterium 69,70 69,70 69,70 75,76 78,79 69,70 Proteobacteria 81,82 66,67 72,73 81,82 75,76 72,73 Gammaproteobacteria 81,82 66,67 72,73 87,88 75,76 72,73 Firmicutes 72,73 75,76 66,67 78,79 69,70 63,64 Clostridia 75,76 69,70 81,82 72,73 66,67 78,79 As seen in Table 5 , when the bacteria called Proteobacteria or Gammaproteobacteria were removed, the accuracy increased in linear regression, MLP and SVM linear models, while the accuracy decreased only in the Naive Bayes model. The effective reason for bacterial elimination is that in the absence of Proteobacteria bacteria, the MLP model showed 81.82% accuracy, while in the absence of Gammaproteobacteria bacteria, the accuracy in the same model was 87.88%. For this reason, the new test was performed by removing Gammaproteobacteria and the results are given in Table 6 . Table 6 Demographic information and results of Test 4 with 5 bacteria. Bacteria Name Extracted from Dataset Model Accuracy (%) Linear Regression Naive Bayes Linear Regression MLP Linear Regression SVM Poly All Data 81,82 66,67 72,73 87,88 75,76 72,73 Enterococcus 78,79 66,67 63,64 78,79 78,79 63,64 Faecalibacterium 72,73 63,64 69,70 78,79 78,79 69,70 Proteobacteria 75,76 63,64 66,67 69,70 75,76 66,67 Firmicutes 75,76 63,64 72,73 72,73 75,76 69,70 Clostridia 78,79 66,67 78,79 75,76 75,76 75,76 When the results obtained with Test 4 are examined, in the absence of Clostridia , the accuracy increased in the decision tree and SVM poly models, and in the absence of Enterococcus or Faecalibacterium , the accuracy increased in the SVM linear model. However, the highest accuracy of 87.88% obtained throughout all tests decreased to 69.70%. For this reason, it was decided not to perform new tests with the fourth test, as it was anticipated that removing other bacterial data from the dataset would further reduce accuracy. The highest accuracy across all tests was 87.88% with the MLP model when demographic information was included in addition to Enterococcus , Faecalibacterium , Proteobacteria , Gammaproteobacteria , Firmicutes and Clostridia bacteria. The second highest accuracy was 84.85% when the SVM linear model used all bacterial information in addition to demographic information. In addition to the tests performed, the effect of demographic information was examined, and four tests were performed again with the logic of removing the bacteria that decreased the accuracy. Demographic information was not used in these tests. As a result of the tests performed, in the presence of all bacteria, linear regression, naive bayes and MLP models achieved a accuracy of 75.76%; decision tree, SVM linear and SVM poly models achieved a accuracy of 69.70%. With the removal of Clostridia bacteria from the dataset, decision tree, SVM linear and SVM poly models achieved the highest accuracy in the test and provided an accuracy rate of 81.82%. However, since the bacteria that decreased the accuracy was removed from the dataset, the first bacteria removed was Proteobacteria . The highest accuracy with 7 bacteria was linear regression and SVM linear model, and the accuracies were 78.79%. In the absence of Enterobacteriaceae or clostridia among these 7 bacteria, the SVM linear model showed the highest accuracy in the test with 84.85%. Then, Gammaproteobacteria bacteria were also removed from the dataset and the test was repeated. In the results obtained with the remaining 6 bacteria in the dataset, the highest accuracy was 81.82% with the SVM linear model. Again, with the removal of Faecelibacterium bacteria on this dataset, linear regression and MLP models provided 81.82% accuracy. As a result, the highest accuracy was 87.88% in the presence of demographic information, while the highest accuracy was 84.85% in the absence of demographic information. In the presence of demographic information, the highest accuracy was achieved with the MLP model, and in the absence of demographic information, with the SVM linear model. While the MLP model required 5 bacterial information to reach a accuracy of 87.88%, the SVM linear model could reach a accuracy of 84.85% with 6 bacterial information. In this case, it was observed that the presence of demographic information slightly increased the accuracy. The taxa removed from our model were cross-referenced with published studies on microbial signatures associated with CRC. Fusobacterium nucleatum , Bacteroides fragilis , Enterococcus spp ., and Clostridia are among the bacteria proven to be associated with CRC (Sameni et al., 2025 ; Jones et al., 2024 ; Shariati et al., 2021 ). The removal of these taxa was conducted to enhance model accuracy while preserving biologically meaningful patterns. Studies in the literature have shown that certain taxa can exhibit high collinearity or introduce data noise, potentially leading to misclassification errors. To prevent such inaccuracies, these organisms may need to be excluded. During the feature selection process, the biological significance of each removed taxon was analyzed. The taxa most strongly correlated with CRC were retained, optimizing the predictive accuracy of the model. The removal of some taxa improved model reliability by preserving biological signals while reducing noise. The MLP and SVM models, which achieved the highest accuracy in the study, demonstrated strong classification performance despite the reduced feature set. Feature elimination not only increased model accuracy but also maintained a consistent representation of CRC-associated microbial profiles. Compared to similar studies, the obtained accuracy values align with known CRC microbiome patterns (Sameni et al., 2025 ; Jones et al., 2024 ). Microbial changes associated with CRC are frequently linked to inflammatory and metabolic processes. The taxa retained in our model were shown to be directly associated with tumor microenvironment, immune response, and CRC progression (Jones et al., 2024 ). The microorganism elimination process has strengthened biological interpretation by preserving critical microbial dynamics involved in CRC development. Discussion CRC is the third most frequently diagnosed cancer type worldwide and ranks second in cancer-related deaths. Emerging machine learning algorithms and artificial intelligence technologies offer opportunities to improve healthcare delivery through big data analytics. The integration of machine learning and metagenomic data represents a significant innovation in CRC research. These technologies can play a critical role in the early diagnosis of cancer and the development of personalized treatment plans. Such biomarkers are important for monitoring disease progression and evaluating treatment response. Including tissue-specific data in the study allows for a more detailed and accurate understanding of CRC. This approach not only improves diagnostic accuracy but also provides critical insights into the tumor microenvironment, aiding in the development of targeted therapies and personalized treatment strategies. The ability to directly examine the interactions between the microbiome associated with cancerous tissue represents a significant advancement over traditional methods that rely solely on nonspecific samples. Additionally, an ANN-based model predicting CRC using global dietary data was shown to misclassify non-CRC phenotypes by 3% and CRC phenotypes by 1% (Abdul Rahman et al., 2023 ). These results demonstrate that ANN models trained on large and heterogeneous datasets can be applied to both young and older adults. The study by Tripathi and colleagues proposes an optimized machine learning model using five different techniques (K-Nearest Neighbors, Decision Trees, Random Forest, Categorical Boosting, and Gaussian Naive Bayes) for classifying histopathological images of colon cancer tissue. The study states that the Categorical Boosting model showed the best performance and was the most suitable approach (Tripathi et al., 2023 ). These findings highlight the potency and effectiveness of machine learning algorithms in CRC diagnosis and biomarker identification. For example, diagnostic models were created using the Random Forest algorithm, which showed higher efficiency when CRC samples were divided into subgroups (Liu et al., 2024 ). Moreover, the analysis of microbiome data using explainable artificial intelligence (XAI) for CRC biomarker identification has been performed. In this context, the Random Forest algorithm has been reported to give the best results in classifying control and CRC patients, with Fusobacterium , Peptostreptococcus , and Parvimonas bacteria associated with CRC (Novielli et al., 2024). Additionally, a study addressing machine learning-based approaches for cancer prediction using microbiome data used Random Forest algorithms to classify five different types of cancer (head and neck, esophagus, stomach, colon, and rectum cancers). The results showed particularly high accuracy scores (over 90%) for colon cancer, although difficulties were encountered in distinguishing esophageal and rectal cancers, which share similarities with colon cancer (Freitas et al., 2023 ). Furthermore, important genomic and microbial biomarkers distinguishing right (RCC) and left colon (LCC) colorectal cancers were identified using Random Forest models. Three different RF models were created, and important features for each model were identified. In the gene model, PRAC1 expression, in the microbial model, the bacteria Ruminococcus gnavus and Clostridium acetireducens , and in the combined model of genes and microbes, PRAC1 expression, HOXC4, Coprococcus eutactus , Ruminococcus gnavus , and Fusobacterium nucleatum were found to be significant (Kolisnik et al., 2023). Additionally, the Random Forest algorithm identified Lachnospiraceae ND3007 group, Escherichia-Shigella , and unclassified Prevotella bacteria as important biomarkers for CRC diagnosis (Lu et al., 2023). Furthermore, in a significant study, the Random Forest model (AUROC: 0.695) showed the best performance for CRC detection, although it was noted that the training time was long and the interpretability was not high. In contrast, the simpler L2-regularized logistic regression model showed similar performance (AUROC: 0.680) with shorter training time and higher interpretability. Machine learning analyses demonstrate the potential of microbial layers as predictive biomarkers for CRC and CRA. Specifically, taxonomically significant biomarkers and microbial functionality (gene families, enzyme reactions, and metabolic pathways) have been associated with CRC (Murovec et al., 2024). These findings indicate that different machine learning techniques and models offer various advantages in CRC diagnosis. In this context, artificial intelligence (AI) techniques, especially the Random Forest algorithm, provide a wide range of applications in CRC diagnosis and treatment. Notably, AI has significant effectiveness in pre-operative, intra-operative, and post-operative phases of CRC treatment. For example, AI assists human classifications during endoscopy, plays a crucial role in intra-operative enhanced image structures and perfusion assessments, and is significant in post-operative monitoring and follow-up (Lingam et al., 2024). These findings also support the role of the gut microbiome in CRC development and the association of inflammation with various bacterial species of the microbiome (Grellier et al., 2024). The revolutionary impact of AI in gut microbiome analysis and interventions has the potential to unravel the complexity of the microbiome and its clinical applications. AI techniques can analyze various microbiome information layers such as metabolomic, transcriptomic, proteomic, and genomic data, revealing significant findings related to human health and diseases. Additionally, AI has significant effects in clinical applications, such as probiotic interventions (Abavisani et al., 2024). Moreover, the microbiome can promote the division, invasion, spread, and survival of cancer cells. Mechanisms such as the activation of tumor signaling cascades, secretion of microbiome-derived functional substances, regulation of mRNA methylation, facilitation of immune evasion by cancer cells, and reshaping of the tumor microecosystem are essential in understanding CRC metastasis (Yu et al., 2023). Stable and non-invasive CRC and adenoma diagnostic models are being developed with the integration of clinical features and cross-cohort metagenomic features. In this context, it has been reported that the AUC of the CRC diagnostic model increased to 0.939 and the AUC of the adenoma diagnostic model to 0.925 with the integration of clinical features (Zhou et al., 2024). Additionally, factors such as transit time, gut inflammation (fecal calprotectin), and body mass index have been found to significantly contribute to CRC diagnostic groups. Fusobacterium nucleatum did not show significant association with CRC diagnostic groups, but Anaerococcus vaginalis , Dialister pneumosintes , Parvimonas micra , Peptostreptococcus anaerobius , Porphyromonas asaccharolytica , and Prevotella intermedia bacteria remained strongly associated (Tito et al., 2024). These findings highlight the importance of microbial factors in CRC diagnosis and suggest that alternative methods in biomarker identification can also be valuable. For example, potential biomarkers for CRC screening and monitoring can be identified using LC-MS (Liquid Chromatography-Mass Spectrometry)-based serum metabolomic analysis. This method has shown that the serum metabolic profile of CRC patients differs significantly from that of the healthy balance group. Additionally, pathways such as arginine biosynthesis, pyrimidine metabolism, pantothenic acid and CoA biosynthesis have been found to be associated with CRC. Nine metabolites have been observed to have significant diagnostic value for CRC, with guanosine having the highest AUC values (Yi et al., 2023). The integration of metabolomic analysis with machine learning algorithms is evidently crucial in CRC diagnosis. Furthermore, using Explainable AI (XAI) for CRC classification based on gut microbiome data, more personalized CRC biomarker identification is performed using the Shapley Additive Explanations (SHAP) method. For instance, in the analysis of five independent datasets, the SHAP method is an effective tool for separating CRC patients into different CRC probability groups and identifying different bacteriome biomarkers in these groups (Rynazal et al., 2023). Models using multiple taxonomic biomarkers have been reported to perform better than those using single taxonomic biomarkers (AUC: 0.83). Additionally, bacterial-fungal coabundance analyses have shown interactions, particularly between Talaromyces islandicus and Clostridium saccharobutylicum bacteria (Liu et al., 2024 ). Finally, in a study using mRNA meta-sequencing analysis to determine differences between the gut microbiomes of healthy individuals and CRC patients, the developed model was validated using independent test datasets and detected CRC from Stage I to Stage II with 60% accuracy. Additionally, no statistically significant differences were observed in true positive rates among samples collected from different cities or different regions of the colorectum (Konishi et al., 2022). These comprehensive findings demonstrate that the Random Forest algorithm and other machine learning techniques are powerful and effective tools in CRC diagnosis and biomarker identification. Furthermore, microbiome analyses play a significant role in this process, providing valuable information for clinical applications. In this context, integrating machine learning (ML) with metagenomic data has opened new pathways to improve diagnostic efficiency and understand the role of the microbiome in CRC. The findings of this study, consistent with previous research, underscore the transformative potential of these approaches. ML and microbiome analyses offer significant advantages not only in CRC diagnosis but also in personalized medicine applications. However, our study has some limitations. The small sample size limits the generalizability of the findings, necessitating larger cohort studies. The collection of metagenomic and tissue-specific data from CRC patients is limited due to ethical regulations and hospital procedures. The process of obtaining biological samples and patient data requires voluntary participation, making data acquisition time-consuming. High-resolution metagenomic analyses are financially expensive, posing challenges for large-scale sampling. Additionally, the computational power required for data processing can restrict working with large datasets. Small datasets can be sufficient for testing specific hypotheses and allow for more detailed biological interpretations. Preliminary studies assess accuracy and biological relevance before transitioning to larger datasets, providing scientific justification for limited sample usage. Most previous CRC studies have utilized similar dataset sizes, and our study remains consistent with the sample sizes used in current biomedical research. Findings from small-scale datasets should be validated through larger studies, and we plan to test our results on larger populations in future phases. The lack of diversity within the study cohort reduces the applicability of the results to broader populations. The data were collected from a single center, and multicenter studies are required to validate the findings. The gut microbiome is influenced by various uncontrollable factors. Future studies should ensure comprehensive ethical review and informed consent. The exclusion of multi-omic data is a significant limitation. Addressing these limitations will help improve ML models and enhance their clinical applicability in CRC diagnosis. Conclusion This study underscores the transformative potential of machine learning (ML) when integrated with metagenomic and tissue-specific data to enhance colorectal cancer (CRC) diagnostics. The inclusion of demographic data further amplifies the predictive capabilities of these models, emphasizing the value of a holistic approach in medical diagnostics. The findings of this study reveal that critical bacterial biomarkers, identified through meticulous feature selection, play a pivotal role in distinguishing CRC patients from healthy controls. These results resonate with previous research that has shown improved diagnostic efficiency of CRC subgroups through ML models based on intestinal microbes. The iterative process of optimizing feature selection by removing less informative bacterial data not only enhances model performance but also sharpens our understanding of the microbiome's intricate role in CRC. This approach provides a pathway to more precise, non-invasive diagnostic tools, potentially revolutionizing CRC detection and monitoring. Future research should aim to validate these findings across larger and more diverse cohorts to ensure broader applicability. Additionally, integrating other data layers such as metabolomic and transcriptomic profiles could further refine these models, paving the way for even greater diagnostic precision and personalized treatment strategies. The application of ML to metagenomic data represents a significant leap forward in CRC diagnostics, offering new insights into the microbiome's role and setting the stage for advanced, data-driven medical solutions. This study not only contributes to the growing body of knowledge but also lays the groundwork for future innovations in cancer diagnosis and treatment. Declarations Acknowledgements We are grateful to the Balcalı hospital institution and the patients for their contributions to this metagenomic study. Authorship Contributions A.D. Y.Ü. F.A. U.O. U.U. E.G. S.D. All authors were involved in study design, analysis and interpretation data. A.D. study conception and design, analysis and interpretation of data, and manuscript outline. F.A. U.O. U.U. E.G. analysis and interpretation of data and manuscript outline. A.D., Y.U. obtaining data and critical revision of the manuscript, analysis and interpretation of data. Funding This study was supported by the Turkish Gastroenterology Association. Data availability Availability of data and materials Genomic sequences used in this study have been uploaded to the NCBI GenBank database. The relevant datasets are publicly available under the following accession numbers (Sequence Read Archive submission: SUB14430300, BioProject ID: PRJNA1112781) Ethics approval and consent to participate This study complied with the ethical guidelines of the Declaration of Helsinki. Approval for this study was received by the Çukurova University Ethics Committee on November 3, 2023 (institutional review board and clinical trial number, IRB numbers: 2023-26-138). Informed consent was obtained from all participants. All authors maintained relevant ethical standards to preserve research integrity and ensured that no duplication, falsification, or plagiarism occurred. Consent for publication Not applicable. Competing interests The authors declare no competing interests. References Abdul Rahman, H., Ottom, M. A., & Dinov, I. D. (2023). Machine learning-based colorectal cancer prediction using global dietary data. BMC cancer, 23(1), 144. Ali, Z. M., Hassoon, N. H., Ahmed, W. S., & Abed, H. N. (2020). The application of data mining for predicting academic performance using k-means clustering and naïve bayes classification. International Journal of Psychosocial Rehabilitation, 24(03), 2143-2151. Anderson, S. M., & Sears, C. L. (2023). Role of the Gut Microbiome in Cancer: A Review, With Special Focus on Colorectal Neoplasia and Clostridioides difficile. Clinical Infectious Diseases, 77(Supplement 6), S471-S478. Arnold, M., Sierra, M. S., Laversanne, M., Soerjomataram, I., Jemal, A., & Bray, F. (2017). Global patterns and trends in colorectal cancer incidence and mortality. Gut, 66(4), 683-691. Baba, Y., Huttenhower, C., Nosho, K., Tanaka, N., Shima, K., Hazra, A., ... & Ogino, S. (2010). Epigenomic diversity of colorectal cancer indicated by LINE-1 methylation in a database of 869 tumors. Molecular cancer, 9, 1-17. Boser, B. E., Guyon, I. M., & Vapnik, V. N. (1992, July). A training algorithm for optimal margin classifiers. In Proceedings of the fifth annual workshop on Computational learning theory (pp. 144-152). Bray, J. R., & Curtis, J. T. (1957). An Ordination of the Upland Forest Communities of Southern Wisconsin. Ecological Monographs , 27 (4). https://doi.org/10.2307/1942268 Chen, S., Webb, G. I., Liu, L., & Ma, X. (2020). A novel selective naïve Bayes algorithm. Knowledge-Based Systems, 192, 105361. Cortes, C. (1995). Support-Vector Networks. Machine Learning. Dekker, E., Tanis, P. J., Vleugels, J. L., Kasi, P. M., & Wallace, M. B. (2019). Colorectal cancer. The Lancet, 394(10207), 1467-1480. Delik, A., Dinçer, S., Ülger, Y., Akkız, H., & Karaoğullarından, Ü. (2022). Metagenomic identification of gut microbiota distribution on the colonic mucosal biopsy samples in patients with non-alcoholic fatty liver disease. Gene , 833 . https://doi.org/10.1016/j.gene.2022.146587 Freitas, P., Silva, F., Sousa, J. V., Ferreira, R. M., Figueiredo, C., Pereira, T., & Oliveira, H. P. (2023). Machine learning-based approaches for cancer prediction using microbiome data. Scientific Reports , 13 (1), 11821. Guyon, I., Boser, B., & Vapnik, V. (1992). Automatic capacity tuning of very large VC-dimension classifiers. Advances in neural information processing systems, 5. Harris, C. R., Millman, K. J., van der Walt, S. J., Gommers, R., Virtanen, P., Cournapeau, D., Wieser, E., Taylor, J., Berg, S., Smith, N. J., Kern, R., Picus, M., Hoyer, S., van Kerkwijk, M. H., Brett, M., Haldane, A., del Río, J. F., Wiebe, M., Peterson, P., … Oliphant, T. E. (2020). Array programming with NumPy. In Nature (Vol. 585, Issue 7825). https://doi.org/10.1038/s41586-020-2649-2 Hunter, J. D. (2007). Matplotlib: A 2D graphics environment. Computing in Science and Engineering , 9 (3). https://doi.org/10.1109/MCSE.2007.55 Jones, J., Shi, Q., Nath, R. R., & Brito, I. L. (2024). Keystone pathobionts associated with colorectal cancer promote oncogenic reprograming. Plos one, 19(2), e0297897. Kho, Z. Y., & Lal, S. K. (2018). The human gut microbiome – A potential controller of wellness and disease. Frontiers in Microbiology, 9, 1835. Liu, G., Su, L., Kong, C., Huang, L., Zhu, X., Zhang, X., ... & Wang, J. (2024). Improved diagnostic efficiency of CRC subgroups revealed using machine learning based on intestinal microbes. BMC gastroenterology , 24 (1), 315. McKinney, W., & Team, P. D. (2015). Pandas-Powerful python data analysis toolkit. Pandas—Powerful Python Data Analysis Toolkit , 1625 . https://doi.org/10.5281/zenodo.3509134 Neuwirth, E. (2014). RColorBrewer: ColorBrewer palettes. R Package Version 1.1-2 , https://cran.R-project.org/package=RColorBrewer. https://doi.org/10.32614/CRAN.package.RColorBrewer Oh, J. H., Lee, J. H., Cho, M. S., Kim, H., Chun, J., Lee, J. H., Yoon, Y., & Kang, W. (2021). Characterization of gut microbiome in Korean patients with metabolic associated fatty liver disease. Nutrients , 13 (3). https://doi.org/10.3390/nu13031013 Ohata, E. F., Chagas, J. V. S. D., Bezerra, G. M., Hassan, M. M., de Albuquerque, V. H. C., & Filho, P. P. R. (2021). A novel transfer learning approach for the classification of histological images of colorectal cancer. The Journal of Supercomputing, 1-26. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M., & Duchesnay, É. (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research , 12 . https://www.jmlr.org/papers/volume12/pedregosa11a/pedregosa11a.pdf?ref=https:/ Perez, J. G., & Perez, E. S. (2021). Predicting student program completion using Naïve Bayes classification algorithm. International Journal of Modern Education and Computer Science, 12(3), 57. Rebersek, M. (2021). Gut microbiome and its role in colorectal cancer. BMC cancer , 21 (1), 1325. Rooks, M. G., & Garrett, W. S. (2016). Gut microbiota, metabolites, and host immunity. Nature Reviews Immunology, 16(6), 341-352. Sameni, F., Elkhichi, P. A., Dadashi, A., Sadeghi, M., Goudarzi, M., Eshkalak, M. P., & Dadashi, M. (2025). Global prevalence of Fusobacterium nucleatum and Bacteroides fragilis in patients with colorectal cancer: an overview of case reports/case series and meta-analysis of prevalence studies. BMC gastroenterology, 25(1), 71. Schölkop, B., Burgest, C., & Vapnik, V. (1995, August). Extracting support data for a given task. In Proceedings, First International Conference on Knowledge Discovery & Data Mining. AAAI Press, Menlo Park, CA (pp. 252-257). Schölkopf, B., Burgest, C., & Vapnik, V. (1996). Incorporating invariances in support vector learning machines. In Artificial Neural Networks—ICANN 96: 1996 International Conference Bochum, Germany, July 16–19, 1996 Proceedings 6 (pp. 47-52). Springer Berlin Heidelberg. Shafi, A. S. M., Molla, M. I., Jui, J. J., & Rahman, M. M. (2020). Detection of colon cancer based on microarray dataset using machine learning as a feature selection and classification techniques. SN Applied Sciences, 2, 1-8. Shafi, M. A., & Rusiman, M. S. (2015). The use of fuzzy linear regression models for tumor size in colorectal cancer in hospital of Malaysia. Applied Mathematical Sciences, 9(56), 2749-2759. Shariati, A., Razavi, S., Ghaznavi-Rad, E., Jahanbin, B., Akbari, A., Norzaee, S., & Darban-Sarokhalil, D. (2021). Association between colorectal cancer and Fusobacterium nucleatum and Bacteroides fragilis bacteria in Iranian patients: a preliminary study. Infectious agents and cancer, 16(1), 41. Siegel, R. L., Miller, K. D., & Jemal, A. (2020). Cancer statistics, 2020. CA: A Cancer Journal for Clinicians, 70(1), 7-30. Spellerberg, I. F., & Fedor, P. J. (2003). A tribute to Claude-Shannon (1916-2001) and a plea for more rigorous use of species richness, species diversity and the “Shannon-Wiener” Index. Global Ecology and Biogeography , 12 (3). https://doi.org/10.1046/j.1466-822X.2003.00015.x Tharwat, M., Sakr, N. A., El-Sappagh, S., Soliman, H., Kwak, K. S., & Elmogy, M. (2022). Colon cancer diagnosis based on machine learning and deep learning: modalities and analysis techniques. Sensors, 22(23), 9250. Tripathi, A., Misra, A., Kumar, K., & Chaurasia, B. K. (2023). Optimized Machine Learning for classifying colorectal tissues. SN Computer Science, 4(5), 461. Vapnik, V. N. (1963). Pattern recognition using generalized portrait method. Automation and remote control, 24(6), 774-780. Vapnik, V., Golowich, S., & Smola, A. (1996). Support vector method for function approximation, regression estimation and signal processing. Advances in neural information processing systems, 9. Virtanen, P., Gommers, R., Oliphant, T. E., Haberland, M., Reddy, T., Cournapeau, D., Burovski, E., Peterson, P., Weckesser, W., Bright, J., van der Walt, S. J., Brett, M., Wilson, J., Millman, K. J., Mayorov, N., Nelson, A. R. J., Jones, E., Kern, R., Larson, E., … Vázquez-Baeza, Y. (2020). SciPy 1.0: fundamental algorithms for scientific computing in Python. Nature Methods , 17 (3). https://doi.org/10.1038/s41592-019-0686-2 Wang, X., Sun, X., Chu, J., Sun, W., Yan, S., & Wang, Y. (2023). Gut microbiota and microbiota-derived metabolites in colorectal cancer: enemy or friend. World Journal of Microbiology and Biotechnology , 39 (11), 291. Wickham, H., Averick, M., Bryan, J., Chang, W., McGowan, L., François, R., Grolemund, G., Hayes, A., Henry, L., Hester, J., Kuhn, M., Pedersen, T., Miller, E., Bache, S., Müller, K., Ooms, J., Robinson, D., Seidel, D., Spinu, V., … Yutani, H. (2019). Welcome to the Tidyverse. Journal of Open Source Software , 4 (43).https://doi.org/10.21105/joss.01686 Zeng, R., Gou, H., Lau, H. C. H., & Yu, J. (2024). Stomach microbiota in gastric cancer development and clinical implications. Gut, 73(12), 2062. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Major Revisions Needed 20 Jan, 2026 Reviewers agreed at journal 27 Nov, 2025 Reviewers invited by journal 25 Nov, 2025 Editor assigned by journal 08 Oct, 2025 First submitted to journal 08 Oct, 2025 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-7808541","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":550527701,"identity":"6bf382e3-a804-4814-96af-4a9fe66aac9a","order_by":0,"name":"Anıl DELİK","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3UlEQVRIiWNgGAWjYFACxsYDPAwWPBLsDUCOgQVRWhqAWiR4JHgOgLRIEGcPSAuDhEQCiE2EFv7ZzQ0H3lRIyEjOfH51w48CCQb+9u4EvFok7hxsODjnjASPtHRO2c0eoMMkzpzdgN+aG4kNh3nbJHjkpHPSbvAAtRhI5OLXIg/W8g+oRfJM2s0/xGgxAGtpADpMgv3YbaJsMQRqOTjnmASPZE8O220ZA2BoE/KL3I30hw/e1NjYSxw//uzmmz82cvztvQS8jwA8BmCSWOUgwP6AFNWjYBSMglEwggAA345Hol+5OJAAAAAASUVORK5CYII=","orcid":"","institution":"Çukurova Universitesi: Cukurova Universitesi","correspondingAuthor":true,"prefix":"","firstName":"Anıl","middleName":"","lastName":"DELİK","suffix":""},{"id":550527702,"identity":"4016262c-dad8-4796-a68b-c96249cfbf06","order_by":1,"name":"Yakup Ülger","email":"","orcid":"","institution":"Çukurova Üniversitesi Tıp Fakültesi: Cukurova Universitesi Tip Fakultesi","correspondingAuthor":false,"prefix":"","firstName":"Yakup","middleName":"","lastName":"Ülger","suffix":""},{"id":550527703,"identity":"acaf5804-ffbf-4402-a117-30db49ed669c","order_by":2,"name":"Ferhat Albayrak","email":"","orcid":"","institution":"Çukurova Universitesi: Cukurova Universitesi","correspondingAuthor":false,"prefix":"","firstName":"Ferhat","middleName":"","lastName":"Albayrak","suffix":""},{"id":550527704,"identity":"3814ff7c-5d1d-4b51-800d-96588994f080","order_by":3,"name":"Umut Orhan","email":"","orcid":"","institution":"Çukurova Universitesi: Cukurova Universitesi","correspondingAuthor":false,"prefix":"","firstName":"Umut","middleName":"","lastName":"Orhan","suffix":""},{"id":550527705,"identity":"3f90004e-64ef-42fc-9205-3dc424217197","order_by":4,"name":"Ulku Unal","email":"","orcid":"","institution":"Adama Science and Technology University","correspondingAuthor":false,"prefix":"","firstName":"Ulku","middleName":"","lastName":"Unal","suffix":""},{"id":550527706,"identity":"e05c3a9c-3b9f-4ab4-bac5-53f4ac20f5d7","order_by":5,"name":"Esra Gov","email":"","orcid":"","institution":"Adana Alparslan Turkes Bilim ve Teknoloji Universitesi","correspondingAuthor":false,"prefix":"","firstName":"Esra","middleName":"","lastName":"Gov","suffix":""},{"id":550527707,"identity":"c455a304-ddc3-43d9-a283-b948fb1a2774","order_by":6,"name":"Sadık Dinçer","email":"","orcid":"","institution":"Cukurova Universitesi","correspondingAuthor":false,"prefix":"","firstName":"Sadık","middleName":"","lastName":"Dinçer","suffix":""}],"badges":[],"createdAt":"2025-10-08 13:49:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7808541/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7808541/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":97257564,"identity":"0268d45a-1448-425e-ad89-604adf7b2bea","added_by":"auto","created_at":"2025-12-02 13:40:56","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":619308,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/f4764dd2f91f186f0ba3daa3.docx"},{"id":97257555,"identity":"e3866634-1475-45ba-a78c-393efcce02ed","added_by":"auto","created_at":"2025-12-02 13:40:56","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":307437,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.docx","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/5aeaa85ebf6c8742cdef83db.docx"},{"id":97366713,"identity":"96ce1bb2-03ca-48e2-9bf9-2a143752c896","added_by":"auto","created_at":"2025-12-03 16:03:36","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":112945,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.docx","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/5bbd0fe6fb58f0ea285589b9.docx"},{"id":97257556,"identity":"367e54e6-f95c-4ac0-89c7-c2319a89aaa4","added_by":"auto","created_at":"2025-12-02 13:40:56","extension":"xml","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10142,"visible":true,"origin":"","legend":"","description":"","filename":"joagJOAGD2500481.xml","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/9f53891418ada06205cf4c61.xml"},{"id":97368079,"identity":"31eaa193-1a3a-4a62-b759-7ad5c841d2f2","added_by":"auto","created_at":"2025-12-03 16:21:32","extension":"xml","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1039,"visible":true,"origin":"","legend":"","description":"","filename":"JOAGD25004818851.go.xml","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/8a589df11fb95a260f23d319.xml"},{"id":97257558,"identity":"018d25ee-5c26-4491-816e-6663ffb307e0","added_by":"auto","created_at":"2025-12-02 13:40:56","extension":"xml","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":802,"visible":true,"origin":"","legend":"","description":"","filename":"JOAGD2500481Import.xml","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/fc35177120a41e10e52cd2e8.xml"},{"id":97257571,"identity":"584c0cd5-396a-4f5d-b996-cf356a2381a7","added_by":"auto","created_at":"2025-12-02 13:40:57","extension":"xml","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":169153,"visible":true,"origin":"","legend":"","description":"","filename":"JOAGD25004810enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/dc96f797e91eb3c4aaf9a069.xml"},{"id":97367002,"identity":"f29e170b-4960-4b6c-8502-01dab58cd7b5","added_by":"auto","created_at":"2025-12-03 16:15:25","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1167,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/d08ea8a93b21bd1c05205d69.png"},{"id":97257573,"identity":"b934caf2-1a22-4027-8b70-33d020b9c7dd","added_by":"auto","created_at":"2025-12-02 13:40:57","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":605065,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/08cbb28a4498a8be1740863c.png"},{"id":97257565,"identity":"871f261d-2e6b-4cef-b11a-967354abeeb3","added_by":"auto","created_at":"2025-12-02 13:40:56","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":293700,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/f0b427c2149e0b9d4b85ff18.png"},{"id":97257563,"identity":"15d637b1-a80a-44cd-87ff-ef3fdb72ffa1","added_by":"auto","created_at":"2025-12-02 13:40:56","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":99192,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/863ded67defeb18276b0ba46.png"},{"id":97257570,"identity":"9e42b45a-8d3e-4b1f-b5ff-b8577fb020e9","added_by":"auto","created_at":"2025-12-02 13:40:57","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":906,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/b7ea765bbb0fd36e01bfae22.png"},{"id":97367267,"identity":"32672fd2-5ff3-4589-812e-592706da067a","added_by":"auto","created_at":"2025-12-03 16:17:55","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":78263,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/1dfea280486dcec21e0eb076.png"},{"id":97257569,"identity":"f75e2ae4-2943-47c3-812b-ae4c304254be","added_by":"auto","created_at":"2025-12-02 13:40:57","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":37917,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/3b6bb15362741fa949a53bd4.png"},{"id":97366736,"identity":"e9e13f7a-eb40-4c9c-9d8b-debc92241992","added_by":"auto","created_at":"2025-12-03 16:04:51","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29475,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/01b581bc7263eb52c16217ff.png"},{"id":97366693,"identity":"339d48c0-0b0c-41b2-aff2-bdba448ab626","added_by":"auto","created_at":"2025-12-03 16:00:02","extension":"xml","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":166706,"visible":true,"origin":"","legend":"","description":"","filename":"JOAGD25004810structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/2b18de59c37d866c449c518d.xml"},{"id":97257566,"identity":"0c94049f-9a95-4fb0-a535-ac30a0043765","added_by":"auto","created_at":"2025-12-02 13:40:57","extension":"html","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":182293,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/c8877423e4608de388daf1be.html"},{"id":97257554,"identity":"cd233d02-8275-43f1-bb35-46e96b2b83a0","added_by":"auto","created_at":"2025-12-02 13:40:56","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":548284,"visible":true,"origin":"","legend":"\u003cp\u003eThe figure presents the relative abundance of bacterial taxa at four taxonomic levels—Phylum, Class, Family, and Genus—in colorectal cancer (CRC) and Control samples. \u003cstrong\u003eA\u003c/strong\u003e Phylum-Level Composition, The relative abundance of \u003cem\u003eFirmicutes\u003c/em\u003eand \u003cem\u003eProteobacteria\u003c/em\u003e is compared between CRC and Control samples. \u003cem\u003eFirmicutes\u003c/em\u003e(light blue) are notably more prevalent in CRC samples, whereas \u003cem\u003eProteobacteria\u003c/em\u003e (dark blue) show a higher abundance in Control samples. Unclassified taxa (beige) remain present in both groups. \u003cstrong\u003eB\u003c/strong\u003e Class-Level Composition, The figure illustrates the distribution of \u003cem\u003eClostridia\u003c/em\u003e(light blue) and \u003cem\u003eGammaproteobacteria\u003c/em\u003e(dark blue) across CRC and Control samples. CRC samples demonstrate an elevated abundance of \u003cem\u003eClostridia\u003c/em\u003e, while Control samples display a greater presence of \u003cem\u003eGammaproteobacteria\u003c/em\u003e. Unclassified bacterial taxa (beige) are observed across both sample groups. \u003cstrong\u003eC\u003c/strong\u003eFamily-Level Composition, \u003cem\u003eEnterobacteriaceae\u003c/em\u003e(dark blue) is highlighted in this panel, showing its relative abundance differences between CRC and Control samples. The data indicate a lower prevalence of \u003cem\u003eEnterobacteriaceae\u003c/em\u003e in CRC compared to Control samples. The presence of unclassified taxa (beige) is recorded in both groups. \u003cstrong\u003eD\u003c/strong\u003e Genus-Level Composition, The figure showcases the differential abundance of three genera: \u003cem\u003eBacteroides\u003c/em\u003e (light blue), \u003cem\u003eEnterococcus\u003c/em\u003e (green), and \u003cem\u003eFaecalibacterium\u003c/em\u003e (dark blue). CRC samples exhibit a higher abundance of \u003cem\u003eBacteroides\u003c/em\u003eand \u003cem\u003eEnterococcus\u003c/em\u003e, whereas \u003cem\u003eFaecalibacterium\u003c/em\u003e is more dominant in Control samples. Unclassified taxa (beige) persist in both conditions.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/3999610d5113d16f96509df7.png"},{"id":97257561,"identity":"53c2c880-f767-4362-9258-da94cfb045db","added_by":"auto","created_at":"2025-12-02 13:40:56","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":255263,"visible":true,"origin":"","legend":"\u003cp\u003eAlpha diversity analysis results by using the Shannon index. \u0026nbsp;Shannon Entropy Index by Group This figure presents the Shannon entropy index (H') across various Control and Colorectal Cancer (CRC) samples, providing insights into the overall microbial diversity within each sample group. The x-axis denotes individual samples, labeled as S15, S16, S17, S18, S20, CA1, CA2, CA3, CA4, CA5, CA6, CA7, CA8, CA9, CA10, CA11, CA12, CA13. The y-axis represents Shannon entropy (H'), measuring the diversity of microbial species within each sample. The scale ranges from 0.0 to 1.0, where higher values indicate greater microbial diversity. Control samples (blue bars) and CRC samples (red bars) are visually distinguished in the figure. p-value of 0.077 is displayed, indicating a statistical comparison between the Shannon entropy indices of the two groups.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/4eeedac5387bc28a27b55f7e.png"},{"id":97367965,"identity":"d4964f5b-9f29-4a6c-80a6-cc2b3c48ce75","added_by":"auto","created_at":"2025-12-03 16:21:08","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":99998,"visible":true,"origin":"","legend":"\u003cp\u003eBeta diversity was visualised by Principal Coordinate Analysis (PCoA). PcoA plot illustrating the distribution of microbiome samples from control (blue) and colorectal cancer (CRC, red) groups. Each point represents an individual sample positioned according to its principal coordinate values, with PCoA1 explaining 34.23% of the variation and PCoA2 accounting for 21.09%. Ellipses denote the clustering patterns within each group, highlighting differences in microbial composition.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/774c16cbad891efe2d29898a.png"},{"id":97664555,"identity":"de597163-987c-4ca4-ae2f-bfbb1421c9f2","added_by":"auto","created_at":"2025-12-08 09:09:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2323019,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7808541/v1/0932ab2b-0e64-431e-bbfd-a09509efa045.pdf"}],"financialInterests":"","formattedTitle":"Machine Learning Integration of Tissue-Specific Metagenomic Signatures for Colorectal Cancer Diagnosis","fulltext":[{"header":"Introduction","content":"\u003cp\u003eColorectal cancer (CRC) is a major public health concern worldwide, characterized by high morbidity and mortality rates (Arnold et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Siegel et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Early detection and accurate diagnosis are pivotal for effective treatment and improved survival rates. Traditional diagnostic methods, though effective, often face limitations in sensitivity and specificity (Dekker et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Therefore, there is a pressing need for innovative approaches that can enhance diagnostic precision and provide a deeper understanding of the disease's underlying mechanisms.\u003c/p\u003e\u003cp\u003eRecent advancements have highlighted the significant role of the gut microbiome in the pathogenesis and progression of CRC (Wang et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The gut microbiota, composed of trillions of microorganisms, plays a crucial role in maintaining intestinal homeostasis, modulating immune responses, and influencing inflammatory processes (Rooks and Garret, 2016; Kho and Lal, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Dysbiosis, or the imbalance in microbial composition, has been linked to the development and progression of various cancers, including CRC (Anderson and Sears, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Zeng et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Specific alterations in microbial communities can lead to chronic inflammation, immune dysregulation, and even direct interactions with tumor cells, promoting carcinogenesis (Rebersek et al., 2021).\u003c/p\u003e\u003cp\u003eLeveraging machine learning (ML) to analyze metagenomic data offers a promising avenue for enhancing CRC diagnostics (Liu et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). ML models can process vast amounts of data, identifying patterns and biomarkers that might be overlooked by traditional statistical methods. In the context of CRC, ML can be employed to analyze both microbiome and tissue-specific data, providing a comprehensive understanding of the host-microbiome interactions and their implications for CRC (Freitas et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThis study aims to investigate the potential of ML-based approaches to enhance diagnostic efficiency in CRC by analyzing metagenomic and tissue-specific data. By employing various ML models, including Linear Regression, Naive Bayes, Multilayer Perceptron (MLP), Support Vector Machine (SVM), and Decision Tree, we seek to identify the most effective methods for distinguishing CRC patients from healthy controls. Additionally, we aim to pinpoint the key bacterial taxa that serve as biomarkers for CRC, thereby contributing to the development of non-invasive diagnostic tools and personalized treatment strategies. We hypothesize that ML models, particularly those incorporating demographic and bacterial information, will significantly improve CRC diagnostic accuracy. The iterative feature selection process, aimed at optimizing model performance by excluding less informative bacterial data, is expected to highlight the most critical bacterial biomarkers. This approach not only enhances our understanding of the microbiome's role in CRC but also provides a robust foundation for future clinical applications.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003eSample Collection\u003c/h2\u003e\n \u003cp\u003eThis study includes tissue samples from patients diagnosed with CRC and healthy individuals who visited the Gastroenterology Clinic at \u0026Ccedil;ukurova University Balcalı Hospital Faculty of Medicine. The samples from both patients and the control group were collected in accordance with ethical committee approval and informed consent obtained from the participants.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eEthics and Consent Approvals\u003c/h3\u003e\n\u003cp\u003eThis research received approval from the Ethics Committee of the \u0026Ccedil;ukurova University Faculty of Medicine on November 3, 2023 (IRB numbers: 2023-26-138). Written consent was obtained from all participants. The study was conducted in alignment with the Declaration of Helsinki and adhered to relevant ethical guidelines.\u003c/p\u003e\n\u003ch3\u003ePatient Selection\u003c/h3\u003e\n\u003cp\u003eCRC patients included in the study were individuals with a histopathologically confirmed diagnosis of CRC, ranging from stage I to IV. Patients were compared with a matched healthy control group based on parameters such as age, gender, and cancer stage. Tissue samples from both groups were collected during colonoscopy procedures.\u003c/p\u003e\n\u003ch3\u003eExclusion Criteria\u003c/h3\u003e\n\u003cp\u003ePreviously Diagnosed Malignant Diseases: Patients who have been previously diagnosed or treated for another malignant disease. Severe Comorbid Conditions: Individuals with severe comorbid conditions such as uncontrolled hypertension, heart failure, or active infection. Medication Use: Patients who have taken immunosuppressive drugs or chemotherapy within the last three months. Factors Affecting Microbiota: Individuals who have used antibiotics, probiotics, or prebiotic supplements within the last six months. Insufficient Sample Quality: Samples that were insufficient or contaminated for DNA isolation or sequencing. Lack of Ethical Approval: Individuals without ethical committee approval or who do not adhere to study protocols. Lack of Informed Consent: Individuals who did not sign the informed consent form. Non-Compliance with Study Protocol: Individuals with demographic characteristics such as age, gender, or stage that do not comply with the study protocol.\u003c/p\u003e\n\u003ch3\u003eMetagenomic Data Analysis\u003c/h3\u003e\n\u003cp\u003eDNA was isolated from the collected tissue samples and prepared for high-resolution metagenomic sequencing. The obtained sequencing data were quality controlled and subsequently used for the identification and genetic profiling of microbial species.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e16S rRNA gene sequencing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe V3-V4 region of the 16S rRNA gene was amplified with 341F (5\u0026rsquo;-CCTACGGGNGGCWGCAG-3\u0026rsquo;) and 805R (5\u0026rsquo;-GACTACHVGGGTATCTAATCC-3\u0026rsquo;) primers. Sequence results were obtained using the Illumina MiSeq\u0026trade; platform. Sequence data were automatically converted to FASTQ format by the MiSeq\u0026trade; instrument. Data analysis was performed using QIIME2 software. Greengenes2 database was used for identification and taxonomic assignment of microbiota sequences.\u003c/p\u003e\n\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003eRelative abundances of microbiome OTUs\u003c/h2\u003e\n \u003cp\u003eRelative abundance plots were generated using the R programming language to show how the specimens varied at which taxonomic level. These plots show the diversity of the samples in terms of percentages, showing the variation of the prominent organisms. The \u0026apos;tidyverse\u0026apos; library (Wickham et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e), which provides a large collection of packages for data manipulation and visualization, and the \u0026apos;RColorBrewer\u0026apos; library (Neuwirth, \u003cspan class=\"CitationRef\"\u003e2014\u003c/span\u003e) which provides color palette options in R, were used to visualize data consisting of the taxonomic levels of phylum, class, family and genus.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eNormalization and Preprocessing of Metagenomic Data\u003c/h3\u003e\n\u003cp\u003eMetagenomic data often contain a high proportion of zero values, so methods were applied to minimize the zero inflation problem. Log Transformation (Log10 or Natural Log): To correct extreme skewness, very small values were adjusted by adding\u0026thinsp;+\u0026thinsp;1 before log transformation. CLR (Centered Log-Ratio) Transformation: Given the compositional nature of metagenomic data, they are not treated as independent variables, so a ratio-based transformation was used. TMM (Trimmed Mean of M-values) Normalization: Applied to generalize abundance data and reduce sample-based differences. RA (Relative Abundance) Transformation: Bacterial abundance was normalized against total abundance to maintain comparability. VST (Variance Stabilizing Transformation): Used to allow an unbiased evaluation of biological differences between CRC and healthy groups. Negative Binomial Model: Tools like EdgeR and DESeq2 were employed to analyze bacterial abundance and balance overdispersion. Z-score Standardization: Applied to adapt bacterial abundances to a standardized scale for better comparability. After transformation, a feature selection process was conducted to identify the most meaningful biomarkers. Recursive Feature Elimination (RFE) and SHAP (Shapley Decomposition) were integrated into the modeling process to provide explainable AI-based feature selection.\u003c/p\u003e\n\u003ch3\u003eAlpha diversity in metagenome analysis\u003c/h3\u003e\n\u003cp\u003eAlpha diversity is a metric that used to assess the diversity of ecological communities and refers to the richness of species found in a given habitat or sampling site (Oh et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). In the alpha diversity analysis, the Shannon indexes (\u003cem\u003eH\u0026rsquo;\u003c/em\u003e) (Spellerberg \u0026amp; Fedor, \u003cspan class=\"CitationRef\"\u003e2003\u003c/span\u003e) were calculated at the genus taxonomic level as a method so that both the richness (i.e. the number of genus) and abundance (i.e. the relative distribution of the number of individuals of genus) of the genus in the samples examined were taken into account. The equation for calculating the Shannon index (\u003cem\u003eH\u0026rsquo;\u003c/em\u003e) is expressed as follows:\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img1764592226.png\" width=\"337\" height=\"151\"\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003ep\u003c/em\u003e\u003csub\u003e\u003cem\u003ei\u003c/em\u003e\u003c/sub\u003e is the proportion of individuals belonging to genus \u003cem\u003ei\u003c/em\u003e in the dataset. After the calculation for each sample, the Shannon entrophy index plot was visualized in Python programming language employing \u0026apos;pandas\u0026apos; (McKinney \u0026amp; Team, \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e) library and \u0026apos;pyplot\u0026apos; module of \u0026apos;matplotlib\u0026apos; library (Hunter, \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e).\u003c/p\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003eBeta diversity in metagenome analysis\u003c/h2\u003e\n \u003cp\u003eBeta-diversity analysis, in which similarities between microbiome pairs are quantified in terms of distance (Delik et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e), was performed using the Bray-Curtis dissimilarity measure (Bray \u0026amp; Curtis, \u003cspan class=\"CitationRef\"\u003e1957\u003c/span\u003e). The relative abundance values of members of the Genus taxonomic level were used to create the distance matrix in the Python programming language. The distance matrix was obtained using the \u0026lsquo;pandas\u0026rsquo;(McKinney \u0026amp; Team, \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e) and \u0026lsquo;numpy\u0026rsquo; libraries (Harris et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e), as well as the \u0026apos;pdist\u0026apos; function which calculates the distances between pairs and \u0026apos;squareform\u0026apos; which converts the compressed distance vector into a square distance matrix in the \u0026lsquo;scipy.spatial.distance\u0026rsquo; library (Virtanen et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). To perform principal coordinate analysis (PCoA) using the distance matrix, the MDS (Multidimensional Scaling) class in the \u0026apos;sklearn.manifold\u0026apos; library (Pedregosa et al., \u003cspan class=\"CitationRef\"\u003e2011\u003c/span\u003e), which is used in the Python programming language to transform high-dimensional data sets into a lower dimensional space, was used and the data set was transformed with the \u0026apos;fit_transform\u0026apos; method. The PCoA plot was visualized using the \u0026apos;matplotlib.pyplot\u0026apos; library (Hunter, \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003eMachine Learning Models\u003c/h2\u003e\n \u003cp\u003eIn the study, 5 different models and variations of these models were used: linear regression, naive bayes, multilayer perceptron (MLP), support vector machine (SVM) and decision tree. Linear Regression is a statistical method that models the relationship between a dependent variable (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y\\)\u003c/span\u003e\u003c/span\u003e) and one or more independent variables (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:x\\)\u003c/span\u003e\u003c/span\u003e). The goal is to find the best linear equation that describes how the dependent variable \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y\\)\u003c/span\u003e\u003c/span\u003e varies with respect to the independent variables \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{1},{x}_{2},\\:{x}_{3},\\dots\\:,{x}_{n}\\)\u003c/span\u003e\u003c/span\u003e. This equation is usually expressed as follows:\u003c/p\u003e\n \u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e$$\\:y=\\:{\\beta\\:}_{0}+{\\beta\\:}_{1}*{x}_{1}+{\\beta\\:}_{2}*{x}_{2}+\\dots\\:+{\\beta\\:}_{n}*{x}_{n}+\\epsilon\\:$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003eWhile the dependent variable \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y\\)\u003c/span\u003e\u003c/span\u003e in (1) represents the estimated or in other words the desired target, the variables \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{1},{x}_{2},\\dots\\:,{x}_{n}\\)\u003c/span\u003e\u003c/span\u003e represent the independent variables. Although \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\beta\\:}_{0}\\)\u003c/span\u003e\u003c/span\u003e is a constant term, the variables \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\beta\\:}_{1},\\:{\\beta\\:}_{2},\\dots\\:,{\\beta\\:}_{n}\\)\u003c/span\u003e\u003c/span\u003e are the regression coefficients of the variables \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{1},{x}_{2},\\dots\\:,{x}_{n}\\)\u003c/span\u003e\u003c/span\u003e and the effect of these variables on the dependent variable \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y\\)\u003c/span\u003e\u003c/span\u003e. The variable \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\epsilon\\:\\)\u003c/span\u003e\u003c/span\u003e at the end is the error term and represents the unexplained random errors in the model.\u003c/p\u003e\n \u003cp\u003eLinear regression assumes that the relationship between the independent variables and the dependent variable is linear and estimates this relationship using the ordinary least squares (OLS) method. Linear regression is widely used in fields such as data analysis, machine learning, health, and economics (Baba et al., \u003cspan class=\"CitationRef\"\u003e2010\u003c/span\u003e; Shafi \u0026amp; Rusiman, \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eNaive Bayes is a probabilistic classification algorithm and was developed based on Bayes\u0026apos; theorem. Bayes\u0026apos; theorem is a theorem that calculates the probability of an event occurring using prior knowledge of the events that occurred before that event. Bayes\u0026apos; theorem is expressed as follows:\u003c/p\u003e\n \u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e$$\\:P\\left(A|B\\right)=\\:\\frac{P\\left(B|A\\right)*P\\left(A\\right)}{P\\left(B\\right)}$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003eThe probability \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:P\\left(A|B\\right)\\)\u003c/span\u003e\u003c/span\u003e in (2) shows the probability of event \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:A\\)\u003c/span\u003e\u003c/span\u003e occurring if event \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:B\\)\u003c/span\u003e\u003c/span\u003e occurs. While \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:P\\left(B|A\\right)\\)\u003c/span\u003e\u003c/span\u003e shows the probability of event \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:B\\)\u003c/span\u003e\u003c/span\u003e occurring if event \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:A\\)\u003c/span\u003e\u003c/span\u003e occurs, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:P\\left(A\\right)\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:P\\left(B\\right)\\)\u003c/span\u003e\u003c/span\u003e show the probabilities of events \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:A\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:B\\)\u003c/span\u003e\u003c/span\u003e occurring.\u003c/p\u003e\n \u003cp\u003eThe Naive Bayes algorithm also calculates probability according to the Bayes theorem given in Eq.\u0026nbsp;(\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). When calculating probability, it is calculated with the assumption that all features are independent of each other. The values obtained at the end of the calculation are the probabilities that the sample belongs to the calculated class. The sample is included in the class with the highest value. The Naive Bayes method is widely used in many fields such as statistics, commerce, health, deep learning, and education (Ali et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; Chen et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; Ohata et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e; Perez \u0026amp; Perez, \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eIn Decision Tree, each branch of tree, which works as a tree structure that makes decisions by dividing data, represents a specific decision or result. There are 3 types of nodes in decision trees: root node, internal node, and leaf nodes. The root node is the starting point of the decision tree and contains the entire dataset. Here, the feature that best separates the data is selected. Internal nodes are nodes that start from the root node and each divide the data according to a feature or situation. A test or decision is made on a feature in each internal node. Leaf nodes are the end points of the decision tree and represent a specific class or value. These nodes contain classes in classification problems and numerical values in regression problems. In addition, there are elements called branches in decision trees that represent the paths extending between nodes. Each branch represents the probability of a decision or result.\u003c/p\u003e\n \u003cp\u003eMultilayer Perceptron (MLP), one of the types of artificial neural networks (ANN), has additional hidden layers in addition to the input and output layers, unlike classical ANNs. MLP is a feedforward ANN model developed to solve the XOR problem without linear separability. The MLP model consists of three-layer types: input layer, hidden layer and output layer. The number of hidden layers can be more than one, and the number of inputs \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{1},{x}_{2},\\dots\\:,{x}_{n}\\)\u003c/span\u003e\u003c/span\u003e and outputs \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{y}_{1},{y}_{2},\\dots\\:,{y}_{n}\\)\u003c/span\u003e\u003c/span\u003e can vary.\u003c/p\u003e\n \u003cp\u003eAs in non-multilayer ANNs, the input layer is the data layer that comes to the MLP cell. The incoming data is transmitted to the middle layer, also known as the hidden layer. While the increase in the number of cells in the hidden layer or layers causes the MLP to learn more accurately, it can also cause it to learn or memorize in a longer time. The output layer processes the information coming from the middle layer and produces the output of the MLP.\u003c/p\u003e\n \u003cp\u003eSince its foundations in the 1960s, Support Vector Machine (SVM) is one of the best-known methods that still provides the best classification success in many applications. The algorithm, which was first developed by Vapnik and Lerner (1963), took its current state with studies at AT\u0026amp;T Bell Laboratory. It has become one of the standard methods in machine learning due to its extraordinary success in many different problems (Boser et al., \u003cspan class=\"CitationRef\"\u003e1992\u003c/span\u003e; Guyon et al., \u003cspan class=\"CitationRef\"\u003e1992\u003c/span\u003e; Cortes, \u003cspan class=\"CitationRef\"\u003e1995\u003c/span\u003e; Sch\u0026ouml;lkopf et al., 1995; Sch\u0026ouml;lkopf et al., \u003cspan class=\"CitationRef\"\u003e1996\u003c/span\u003e; Vapnik et al., \u003cspan class=\"CitationRef\"\u003e1996\u003c/span\u003e). The most important feature of SVM is support vectors. While classifying data as a SVM model, it focuses on the most critical points of the data and calls these points support vectors and uses them to define the hyperplane. The selected support vectors are the points closest to the hyperplane. Another important feature is kernel functions. Using these functions, data can be transformed into different spaces. The most used kernel functions are linear, polynomial, radial basis function (RBF) and sigmoid kernel functions.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003eRationale for the Selection of Eight Bacterial Features\u003c/h2\u003e\n \u003cp\u003ePrior to analysis, the dataset contained a total of 672 bacterial species. In the initial phase, all bacterial features were considered; however, a feature selection process was applied, taking into account high collinearity, low variance, and weak biological relevance, which could negatively impact model performance. The feature selection focused not only on classification accuracy but also on preserving taxa that would strengthen biological interpretation. Bacteria with a strongly demonstrated association with CRC in the literature were prioritized during analysis to minimize the risk of excluding key diagnostic patterns from the microbiome. Multiple biological and statistical criteria were considered in the identification of the eight selected bacteria: A literature review of published metagenomic studies and CRC biomarkers was conducted. The distribution of bacterial abundance was statistically analyzed to identify CRC-specific bacteria. Expert opinions on microbiome-CRC relationships were gathered to select features that would enhance diagnostic models.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003eStatistical Analysis\u003c/h2\u003e\n \u003cp\u003eAll statistical analyses were carefully designed to ensure the accurate evaluation of machine learning models and data. The methods used in the analysis of demographic and bacterial data, along with their implementation details, are presented below:\u003c/p\u003e\n \u003cp\u003eBetween-Group Comparisons: To examine the differences in demographic data (age, BMI) between CRC patients and the healthy control group, appropriate statistical tests were used: Age and BMI: Since the data did not meet the assumption of normal distribution, an independent two-sample t-test was not applied to compare mean differences between the two groups. Instead, the non-parametric Mann-Whitney U test was used. Gender: The chi-square test was applied to evaluate the relationship between two categorical variables (group and gender). Given the small sample size, Yates\u0026apos; continuity correction was performed, and its impact on the p-value was reported. Statistical Assumptions: The assumptions for tests such as the t-test and chi-square test were assessed: T-Test: Normal distribution assumptions were evaluated using the Shapiro-Wilk test, and variance homogeneity was tested using Levene\u0026rsquo;s test. Chi-Square Test: Yates\u0026apos; continuity correction was applied to the 2x2 contingency table, and p-values were reported with the correction applied.\u003c/p\u003e\n \u003cp\u003eSummary Statistics:\u003c/p\u003e\n \u003cp\u003eAge and BMI: Reported as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD). Gender distribution: Presented as percentages. Interpretation of P-Values: Hypothesis test p-values were interpreted based on a 5% significance level (\u0026alpha;\u0026thinsp;=\u0026thinsp;0.05). A 95% confidence interval was used in all analyses to ensure more precise comparisons between groups. Two-tailed test results were used, with no preference for one-tailed tests. Bacterial Data Analysis: Due to common metagenomic data characteristics such as skewness, zero inflation, and overdispersion, bacterial data were transformed and normalized. These transformations were performed to meet linear model assumptions and improve model performance. The comparative performance of models was evaluated before and after transformation. Machine Learning Performance Metrics: To ensure class imbalance (13 CRC vs. 20 control) did not affect model performance, additional metrics besides accuracy were used, including AUC-ROC and precision-recall scores, which provided a more comprehensive evaluation of the model\u0026rsquo;s predictive performance. Software and Tools: All statistical analyses were conducted using SPSS software. PCoA (Principal Coordinate Analysis): Bacterial data visualization was performed using Python and relevant libraries (scikit-bio, matplotlib).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003eDemographic feature of study population\u003c/h2\u003e\u003cp\u003eThe average age of CRC patients was 64.83 years with a standard deviation of 13.51, while the control group had an average age of 60.70 years with a standard deviation of 13.47. The difference was not statistically significant (P value\u0026thinsp;=\u0026thinsp;0.20). In the CRC group, 23% were female and 77% were male. In contrast, the control group had an equal distribution of 50% for both genders. The difference in gender distribution had a P value of 0.09, indicating a trend towards significance but not statistically conclusive. The average BMI for the CRC group was 25.85 kg/m\u0026sup2;, and for the control group, it was 24.85 kg/m\u0026sup2;. The difference was not statistically significant (P value\u0026thinsp;=\u0026thinsp;0.28). The majority of tumors in the CRC group were located in the rectum (53.84%), followed by the sigmoid colon (15.38%), ascending colon (15.38%), transversal colon (7.69%), and cecum (7.69%). No patients were at stage T1, 23% were at stage T2, 31% at T3, and 46% at T4, indicating a higher prevalence of advanced-stage tumors. Only 5% of the CRC cases showed invasion, while the vast majority (95%) did not (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\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\u003eDemographic data in colorectal cancer and control individuals\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=\"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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVariable\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCRC (n\u0026thinsp;=\u0026thinsp;13)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eControl (n\u0026thinsp;=\u0026thinsp;20)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e95%(CI)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge (years),( \u003cem\u003emean, SD\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e64.83\u0026thinsp;\u0026plusmn;\u0026thinsp;13.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e60.70\u0026thinsp;\u0026plusmn;\u0026thinsp;13.47\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e(56.76\u0026ndash;72.90)*\u003c/p\u003e\u003cp\u003e(54.22\u0026ndash;67.18)**\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.40\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGender (n,%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e0.146\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3 (23)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e11 (55)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e(6.1\u0026ndash;54.0)*\u003c/p\u003e\u003cp\u003e(32.2\u0026ndash;75.6)**\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10 (77)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e9 (45)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e(46.0\u0026ndash;93.9)*\u003c/p\u003e\u003cp\u003e(24.4\u0026ndash;67.8)**\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBMI (kg/m2), (\u003cem\u003emean, SD\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e25.85\u0026thinsp;\u0026plusmn;\u0026thinsp;2.09\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e24.85\u0026thinsp;\u0026plusmn;\u0026thinsp;5.51\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e(24.46\u0026ndash;27.24)*\u003c/p\u003e\u003cp\u003e(22.14\u0026ndash;27.56)**\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.28\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTumor location (n,%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\" morerows=\"5\" rowspan=\"6\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRectum\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7 (53.84)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eSigmoid colon\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2 (15.38)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eAscending colon\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2 (15.38)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eTransversial colon\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1 (7.69)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eCechum\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1 (7.69)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eTumor staging (T) (n,%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\" morerows=\"4\" rowspan=\"5\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0 (0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eT2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3 (23)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eT3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e4 (31)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eT4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e6 (46)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eİnvasion (n;%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1 (5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\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\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e19 (95)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003eBMI: Body mass index, SD: Standard deviation. CRC, Control. Confidence Intervals (95% CI): Calculated for age, BMI, and gender ratios and included in the table. Used to highlight variability and uncertainty in small sample sizes. Statistical Tests: a: Parametric t-test for age. c: Mann-Whitney test for BMI. b: Chi-square test for gender variable, with Yates' continuity correction applied due to small sample size. P-values: All p-values were assessed bilaterally and reported according to the appropriate tests. Tumor Location and Staging Data: As these data are not applicable for the control group, they were only reported for the CRC group.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003eMicrobiome diversity in samples\u003c/h2\u003e\u003cp\u003eAccording to the analyzed data, relative abundance plots were obtained for the four taxonomic levels, namely phylum, class, family, and genus (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cem\u003eFirmicutes\u003c/em\u003e, which stand out at the phylum level, had a mean value of 38.07% while \u003cem\u003eProteobacteria\u003c/em\u003e had a mean value of 23.11% in total samples. At the class level, \u003cem\u003eGammaproteobacteria\u003c/em\u003e had a mean value of 35.12% and \u003cem\u003eClostridia\u003c/em\u003e had a mean value of 22.49% in total samples. At the family level, \u003cem\u003eEnterobacteriaceae\u003c/em\u003e accounted for 32.32% of the total samples. At the genus level, \u003cem\u003eBacteroides\u003c/em\u003e, \u003cem\u003eEnterococcus\u003c/em\u003e, and \u003cem\u003eFaecalibacterium\u003c/em\u003e had mean values of 12.14%, 5.14%, and 2.29%, respectively. The mean values of the microbiomes in cancer and control samples are given in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e for each taxonomic level.\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\u003eProportional distribution of microbiomes in cancer and control 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=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTaxonomic level\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTaxon name\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMean (Cancer)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eMean (Control)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003ePhylum\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eFirmicutes\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e23.109\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e45.531\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eProteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e37.350\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e38.553\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eClass\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eGammaproteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e21.079\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e44.257\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eClostridia\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e24.237\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e21.368\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFamily\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eEnterobacteriaceae\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e19.744\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e40.502\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eGenus\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eBacteroides\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e19.701\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e7.234\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eEnterococcus\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e10.640\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e1.565\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eFaecalibacterium\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.510\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e2.797\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=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003eAlpha diversity of microbiomes\u003c/h2\u003e\u003cp\u003eBased on the alpha diversity analysis performed at the genus level to evaluate the diversity in the samples, cancer samples (CA13, CA10, and CA11) were determined to have the highest Shannon entropy index values, with 1.021, 0.840, and 0.820, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The lowest alpha diversity was identified in S4 and S9 with an H' index value of 0, and in S7 with an H' index value of 0.012. Mann-Whitney U test was conducted for statistical analysis, yielding a p-value of 0.077. This result indicates that the differences in entropy between the two groups are not statistically significant.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003eBeta diversity of microbiomes\u003c/h2\u003e\u003cp\u003eBeta diversity was visualised by PcoA (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Beta diversity analysis showed that cancer and control samples could be distinguished as two distinct groups. It could be said that sample heterogeneity also had a greater effect on the overall microbiome composition in our samples.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003eMachine learning model results with metagenomic data\u003c/h2\u003e\u003cp\u003eThere are 33 samples in the study, 20 samples in healthy group and 13 samples in control group. Each sample has 10 features, 2 demographic and 8 bacterial information. Demographic information includes age and gender, bacterial information includes \u003cem\u003eBacteriodes\u003c/em\u003e, \u003cem\u003eEnterecoccus\u003c/em\u003e, \u003cem\u003eFaecelibacterium\u003c/em\u003e, \u003cem\u003eProteobacteria\u003c/em\u003e, \u003cem\u003eGammaproteobacteria\u003c/em\u003e, \u003cem\u003eFirmicutes\u003c/em\u003e, \u003cem\u003eEnterobacteriaceae\u003c/em\u003e and \u003cem\u003eClostridia\u003c/em\u003e. These features were selected according to the expert opinion according to the fullness rate, the accuracy of the data and their importance for CRC.\u003c/p\u003e\u003cp\u003eIn the study, 5 different models were used: linear regression, naive bayes, multilayer perceptron, support vector machine and decision tree. For MLP, the hidden layer, the number of neurons in the hidden layer and the optimization method were changed and tested and optimized. For SVM, tests were performed with different kernel functions. In all tests performed;\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eGender and class information were converted to numerical values ​​as 0 (gender for female, class for healthy group) and 1 (gender for male, class for CRC group).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eDue to the small number of samples, it was aimed to increase learning and prediction accuracy by using the leave-one-out cross validation method.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eTests were carried out on all features, and as a result of each test, the bacteria that had a negative effect on the most models were removed and the highest accuracy was tried to be achieved with the least bacterial data.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eThe tests were repeated with and without demographic information in the features and the effect of demographic information was examined.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eIn the tests performed with linear regression, calculations were made by giving all the features and if the results obtained as a result of the regression were equal to or greater than 0.5, it was included in the control group, and if it was smaller, it was included in the healthy group.\u003c/p\u003e\u003cp\u003eStudies conducted with Naive Bayes and Decision Tree methods were conducted without any pre- and post-processing. The accuracy obtained with Naive Bayes were compared with the results of other models as an anchor. It was aimed that other methods, especially deep learning methods, would show higher accuracy than the Naive Bayes method.\u003c/p\u003e\u003cp\u003eIn MLP tests, firstly, the settings that achieved the highest accuracy were selected by using the entire dataset and changing the optimization method, the number of hidden layers, the number of neurons in the hidden layer, and the maximum number of iterations. As a result of the preliminary tests, it was seen that the optimization method adam, the number of hidden layers and neurons as 1 hidden layer 100 neurons, and the maximum iteration as 500 iterations gave the highest accuracy. The settings of the MLP method were determined in this way in the study.\u003c/p\u003e\u003cp\u003ePreliminary tests were performed on the kernel functions, which are the most important feature of the SVM method, using the entire dataset. Linear, polynomial, sigmoid and RBF functions were used as kernel functions. Since the accuracy of the sigmoid and RBF kernel functions in the obtained results remained below the desired level, it was deemed appropriate to perform tests only with linear and polynomial kernel functions in the SVM method.\u003c/p\u003e\u003cp\u003eIn the study, tests were conducted with and without demographic information using 5 models and 6 methods. As a result of these tests, it was observed that demographic information increased the accuracy in estimation. All tests were started with 8 bacteria information and by removing the bacteria information that decreased or did not increase the accuracy at each step, it was aimed to reach the highest accuracy with the least bacteria information. The first test with 8 bacteria and 2 demographic information is given in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\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\u003eDemographic information and results of Test 1 with the entire dataset.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\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=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eBacteria Name Extracted from Dataset\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"6\" nameend=\"c7\" namest=\"c2\"\u003e\u003cp\u003eModel Accuracy (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNaive Bayes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eDecision Tree\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMLP\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eSVM Poly\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAll Data\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e78,79\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e66,67\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e66,67\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u003cb\u003e66,67\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eBacteriodetes\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eEnterococcus\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e57,58\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e54,55\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e51,52\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e60,61\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eFaecalibacterium\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e84,85\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eProteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e81,82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eGammaproteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eFirmicutes\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e60,61\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eEnterobacteriaceae\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eClostridia\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e75,76\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\u003eWhen Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e is examined, it is seen that the Naive Bayes method reached the highest accuracy value with 78.79% when there is demographic information and 8 bacterial information in the dataset. However, since the aim of the study was to reach the highest accuracy with the least bacteria information, each bacterium was removed from the dataset one by one, and the tests were repeated. In this case, when bacteriodetes was removed from the dataset, it was seen that the accuracy increased in linear regression, decision tree, SVM linear and SVM poly models; when \u003cem\u003eFaecalibacterium\u003c/em\u003e or \u003cem\u003eClostridia\u003c/em\u003e was removed from the dataset, it was seen that the accuracy increased in decision tree, MLP, SVM linear and SVM poly models; when proteobacteria was removed from the dataset, it was seen that the accuracy increased in linear regression, MLP and SVM linear models; when \u003cem\u003eGammaproteobacteria\u003c/em\u003e was removed from the dataset, it was seen that the accuracy increased in linear regression and MLP models; when \u003cem\u003eEnterobacteriaceae\u003c/em\u003e was removed from the dataset, it was seen that the accuracy increased in linear regression, decision tree, MLP, SVM linear and SVM poly models. Since the absence of \u003cem\u003eEntreobacteriaceae\u003c/em\u003e data caused the estimation to be more accurate in the 5 models, \u003cem\u003eEntreobacteriaceae\u003c/em\u003e was removed from the dataset and the second test was performed. The results of Test 2 are given in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\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 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eDemographic information and results of Test 2 with 7 bacteria.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\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=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eBacteria Name Extracted from Dataset\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"6\" nameend=\"c7\" namest=\"c2\"\u003e\u003cp\u003eModel Accuracy (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNaive Bayes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMLP\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eSVM Poly\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAll Data\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e75,76\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e75,76\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e69,70\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e75,76\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u003cb\u003e69,70\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eBacteriodetes\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eEnterococcus\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eFaecalibacterium\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eProteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eGammaproteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eFirmicutes\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eClostridia\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e72,73\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\u003eAs in Test 1, there are three bacteria in Test 2 that, when removed from the dataset, increase accuracy on the most models. Among these bacteria, the removal of the bacteria named \u003cem\u003eFaecalibacterium\u003c/em\u003e from the dataset caused an increase in accuracy in two models, while it caused a decrease in accuracy in linear regression, decision tree and SVM poly models. The removal of the bacteria named \u003cem\u003eClostridia\u003c/em\u003e from the dataset caused an increase in two models, while it caused a decrease in accuracy in linear regression and SVM linear models. Finally, the removal of the bacteria named \u003cem\u003eBacteriodetes\u003c/em\u003e caused an increase in accuracy in two models, while it caused a decrease in accuracy in only one model. For this reason, \u003cem\u003eBacteriodetes\u003c/em\u003e was removed from the new dataset containing 7 bacterial information and the third test was performed. The results obtained with Test 3 are given in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\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 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eDemographic information and results of Test 3 with 6 bacteria.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\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=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eBacteria Name Extracted from Dataset\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"6\" nameend=\"c7\" namest=\"c2\"\u003e\u003cp\u003eModel Accuracy (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNaive Bayes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMLP\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eSVM Poly\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAll Data\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e75,76\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e75,76\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eEnterococcus\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eFaecalibacterium\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eProteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e81,82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e81,82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eGammaproteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e81,82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e87,88\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eFirmicutes\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eClostridia\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e81,82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e78,79\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\u003eAs seen in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, when the bacteria called \u003cem\u003eProteobacteria\u003c/em\u003e or \u003cem\u003eGammaproteobacteria\u003c/em\u003e were removed, the accuracy increased in linear regression, MLP and SVM linear models, while the accuracy decreased only in the Naive Bayes model. The effective reason for bacterial elimination is that in the absence of \u003cem\u003eProteobacteria\u003c/em\u003e bacteria, the MLP model showed 81.82% accuracy, while in the absence of \u003cem\u003eGammaproteobacteria\u003c/em\u003e bacteria, the accuracy in the same model was 87.88%. For this reason, the new test was performed by removing \u003cem\u003eGammaproteobacteria\u003c/em\u003e and the results are given in Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\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 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eDemographic information and results of Test 4 with 5 bacteria.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\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=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eBacteria Name Extracted from Dataset\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"6\" nameend=\"c7\" namest=\"c2\"\u003e\u003cp\u003eModel Accuracy (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNaive Bayes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMLP\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLinear Regression\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eSVM Poly\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAll Data\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e81,82\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003e66,67\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e87,88\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e75,76\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u003cb\u003e72,73\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eEnterococcus\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eFaecalibacterium\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eProteobacteria\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eFirmicutes\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e63,64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e72,73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e69,70\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eClostridia\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e66,67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e78,79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e75,76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e75,76\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\u003eWhen the results obtained with Test 4 are examined, in the absence of \u003cem\u003eClostridia\u003c/em\u003e, the accuracy increased in the decision tree and SVM poly models, and in the absence of \u003cem\u003eEnterococcus\u003c/em\u003e or \u003cem\u003eFaecalibacterium\u003c/em\u003e, the accuracy increased in the SVM linear model. However, the highest accuracy of 87.88% obtained throughout all tests decreased to 69.70%. For this reason, it was decided not to perform new tests with the fourth test, as it was anticipated that removing other bacterial data from the dataset would further reduce accuracy.\u003c/p\u003e\u003cp\u003eThe highest accuracy across all tests was 87.88% with the MLP model when demographic information was included in addition to \u003cem\u003eEnterococcus\u003c/em\u003e, \u003cem\u003eFaecalibacterium\u003c/em\u003e, \u003cem\u003eProteobacteria\u003c/em\u003e, \u003cem\u003eGammaproteobacteria\u003c/em\u003e, \u003cem\u003eFirmicutes\u003c/em\u003e and \u003cem\u003eClostridia\u003c/em\u003e bacteria. The second highest accuracy was 84.85% when the SVM linear model used all bacterial information in addition to demographic information.\u003c/p\u003e\u003cp\u003eIn addition to the tests performed, the effect of demographic information was examined, and four tests were performed again with the logic of removing the bacteria that decreased the accuracy. Demographic information was not used in these tests. As a result of the tests performed, in the presence of all bacteria, linear regression, naive bayes and MLP models achieved a accuracy of 75.76%; decision tree, SVM linear and SVM poly models achieved a accuracy of 69.70%. With the removal of \u003cem\u003eClostridia\u003c/em\u003e bacteria from the dataset, decision tree, SVM linear and SVM poly models achieved the highest accuracy in the test and provided an accuracy rate of 81.82%. However, since the bacteria that decreased the accuracy was removed from the dataset, the first bacteria removed was \u003cem\u003eProteobacteria\u003c/em\u003e. The highest accuracy with 7 bacteria was linear regression and SVM linear model, and the accuracies were 78.79%. In the absence of \u003cem\u003eEnterobacteriaceae\u003c/em\u003e or clostridia among these 7 bacteria, the SVM linear model showed the highest accuracy in the test with 84.85%. Then, \u003cem\u003eGammaproteobacteria\u003c/em\u003e bacteria were also removed from the dataset and the test was repeated. In the results obtained with the remaining 6 bacteria in the dataset, the highest accuracy was 81.82% with the SVM linear model. Again, with the removal of \u003cem\u003eFaecelibacterium\u003c/em\u003e bacteria on this dataset, linear regression and MLP models provided 81.82% accuracy.\u003c/p\u003e\u003cp\u003eAs a result, the highest accuracy was 87.88% in the presence of demographic information, while the highest accuracy was 84.85% in the absence of demographic information. In the presence of demographic information, the highest accuracy was achieved with the MLP model, and in the absence of demographic information, with the SVM linear model. While the MLP model required 5 bacterial information to reach a accuracy of 87.88%, the SVM linear model could reach a accuracy of 84.85% with 6 bacterial information. In this case, it was observed that the presence of demographic information slightly increased the accuracy.\u003c/p\u003e\u003cp\u003eThe taxa removed from our model were cross-referenced with published studies on microbial signatures associated with CRC. \u003cem\u003eFusobacterium nucleatum\u003c/em\u003e, \u003cem\u003eBacteroides fragilis\u003c/em\u003e, \u003cem\u003eEnterococcus spp\u003c/em\u003e., and \u003cem\u003eClostridia\u003c/em\u003e are among the bacteria proven to be associated with CRC (Sameni et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Jones et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Shariati et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The removal of these taxa was conducted to enhance model accuracy while preserving biologically meaningful patterns. Studies in the literature have shown that certain taxa can exhibit high collinearity or introduce data noise, potentially leading to misclassification errors. To prevent such inaccuracies, these organisms may need to be excluded. During the feature selection process, the biological significance of each removed taxon was analyzed. The taxa most strongly correlated with CRC were retained, optimizing the predictive accuracy of the model. The removal of some taxa improved model reliability by preserving biological signals while reducing noise. The MLP and SVM models, which achieved the highest accuracy in the study, demonstrated strong classification performance despite the reduced feature set. Feature elimination not only increased model accuracy but also maintained a consistent representation of CRC-associated microbial profiles. Compared to similar studies, the obtained accuracy values align with known CRC microbiome patterns (Sameni et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Jones et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Microbial changes associated with CRC are frequently linked to inflammatory and metabolic processes. The taxa retained in our model were shown to be directly associated with tumor microenvironment, immune response, and CRC progression (Jones et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The microorganism elimination process has strengthened biological interpretation by preserving critical microbial dynamics involved in CRC development.\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eCRC is the third most frequently diagnosed cancer type worldwide and ranks second in cancer-related deaths. Emerging machine learning algorithms and artificial intelligence technologies offer opportunities to improve healthcare delivery through big data analytics. The integration of machine learning and metagenomic data represents a significant innovation in CRC research. These technologies can play a critical role in the early diagnosis of cancer and the development of personalized treatment plans. Such biomarkers are important for monitoring disease progression and evaluating treatment response. Including tissue-specific data in the study allows for a more detailed and accurate understanding of CRC. This approach not only improves diagnostic accuracy but also provides critical insights into the tumor microenvironment, aiding in the development of targeted therapies and personalized treatment strategies. The ability to directly examine the interactions between the microbiome associated with cancerous tissue represents a significant advancement over traditional methods that rely solely on nonspecific samples.\u003c/p\u003e\u003cp\u003eAdditionally, an ANN-based model predicting CRC using global dietary data was shown to misclassify non-CRC phenotypes by 3% and CRC phenotypes by 1% (Abdul Rahman et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These results demonstrate that ANN models trained on large and heterogeneous datasets can be applied to both young and older adults. The study by Tripathi and colleagues proposes an optimized machine learning model using five different techniques (K-Nearest Neighbors, Decision Trees, Random Forest, Categorical Boosting, and Gaussian Naive Bayes) for classifying histopathological images of colon cancer tissue. The study states that the Categorical Boosting model showed the best performance and was the most suitable approach (Tripathi et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These findings highlight the potency and effectiveness of machine learning algorithms in CRC diagnosis and biomarker identification. For example, diagnostic models were created using the Random Forest algorithm, which showed higher efficiency when CRC samples were divided into subgroups (Liu et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Moreover, the analysis of microbiome data using explainable artificial intelligence (XAI) for CRC biomarker identification has been performed. In this context, the Random Forest algorithm has been reported to give the best results in classifying control and CRC patients, with \u003cem\u003eFusobacterium\u003c/em\u003e, \u003cem\u003ePeptostreptococcus\u003c/em\u003e, and \u003cem\u003eParvimonas\u003c/em\u003e bacteria associated with CRC (Novielli et al., 2024). Additionally, a study addressing machine learning-based approaches for cancer prediction using microbiome data used Random Forest algorithms to classify five different types of cancer (head and neck, esophagus, stomach, colon, and rectum cancers). The results showed particularly high accuracy scores (over 90%) for colon cancer, although difficulties were encountered in distinguishing esophageal and rectal cancers, which share similarities with colon cancer (Freitas et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Furthermore, important genomic and microbial biomarkers distinguishing right (RCC) and left colon (LCC) colorectal cancers were identified using Random Forest models. Three different RF models were created, and important features for each model were identified. In the gene model, PRAC1 expression, in the microbial model, the bacteria \u003cem\u003eRuminococcus gnavus\u003c/em\u003e and \u003cem\u003eClostridium acetireducens\u003c/em\u003e, and in the combined model of genes and microbes, PRAC1 expression, HOXC4, \u003cem\u003eCoprococcus eutactus\u003c/em\u003e, \u003cem\u003eRuminococcus gnavus\u003c/em\u003e, and \u003cem\u003eFusobacterium nucleatum\u003c/em\u003e were found to be significant (Kolisnik et al., 2023). Additionally, the Random Forest algorithm identified \u003cem\u003eLachnospiraceae\u003c/em\u003e ND3007 group, \u003cem\u003eEscherichia-Shigella\u003c/em\u003e, and unclassified \u003cem\u003ePrevotella\u003c/em\u003e bacteria as important biomarkers for CRC diagnosis (Lu et al., 2023). Furthermore, in a significant study, the Random Forest model (AUROC: 0.695) showed the best performance for CRC detection, although it was noted that the training time was long and the interpretability was not high. In contrast, the simpler L2-regularized logistic regression model showed similar performance (AUROC: 0.680) with shorter training time and higher interpretability. Machine learning analyses demonstrate the potential of microbial layers as predictive biomarkers for CRC and CRA. Specifically, taxonomically significant biomarkers and microbial functionality (gene families, enzyme reactions, and metabolic pathways) have been associated with CRC (Murovec et al., 2024). These findings indicate that different machine learning techniques and models offer various advantages in CRC diagnosis. In this context, artificial intelligence (AI) techniques, especially the Random Forest algorithm, provide a wide range of applications in CRC diagnosis and treatment. Notably, AI has significant effectiveness in pre-operative, intra-operative, and post-operative phases of CRC treatment. For example, AI assists human classifications during endoscopy, plays a crucial role in intra-operative enhanced image structures and perfusion assessments, and is significant in post-operative monitoring and follow-up (Lingam et al., 2024). These findings also support the role of the gut microbiome in CRC development and the association of inflammation with various bacterial species of the microbiome (Grellier et al., 2024). The revolutionary impact of AI in gut microbiome analysis and interventions has the potential to unravel the complexity of the microbiome and its clinical applications. AI techniques can analyze various microbiome information layers such as metabolomic, transcriptomic, proteomic, and genomic data, revealing significant findings related to human health and diseases. Additionally, AI has significant effects in clinical applications, such as probiotic interventions (Abavisani et al., 2024). Moreover, the microbiome can promote the division, invasion, spread, and survival of cancer cells. Mechanisms such as the activation of tumor signaling cascades, secretion of microbiome-derived functional substances, regulation of mRNA methylation, facilitation of immune evasion by cancer cells, and reshaping of the tumor microecosystem are essential in understanding CRC metastasis (Yu et al., 2023). Stable and non-invasive CRC and adenoma diagnostic models are being developed with the integration of clinical features and cross-cohort metagenomic features. In this context, it has been reported that the AUC of the CRC diagnostic model increased to 0.939 and the AUC of the adenoma diagnostic model to 0.925 with the integration of clinical features (Zhou et al., 2024). Additionally, factors such as transit time, gut inflammation (fecal calprotectin), and body mass index have been found to significantly contribute to CRC diagnostic groups. \u003cem\u003eFusobacterium nucleatum\u003c/em\u003e did not show significant association with CRC diagnostic groups, but \u003cem\u003eAnaerococcus vaginalis\u003c/em\u003e, \u003cem\u003eDialister pneumosintes\u003c/em\u003e, \u003cem\u003eParvimonas micra\u003c/em\u003e, \u003cem\u003ePeptostreptococcus anaerobius\u003c/em\u003e, \u003cem\u003ePorphyromonas asaccharolytica\u003c/em\u003e, and \u003cem\u003ePrevotella intermedia\u003c/em\u003e bacteria remained strongly associated (Tito et al., 2024). These findings highlight the importance of microbial factors in CRC diagnosis and suggest that alternative methods in biomarker identification can also be valuable. For example, potential biomarkers for CRC screening and monitoring can be identified using LC-MS (Liquid Chromatography-Mass Spectrometry)-based serum metabolomic analysis. This method has shown that the serum metabolic profile of CRC patients differs significantly from that of the healthy balance group. Additionally, pathways such as arginine biosynthesis, pyrimidine metabolism, pantothenic acid and CoA biosynthesis have been found to be associated with CRC. Nine metabolites have been observed to have significant diagnostic value for CRC, with guanosine having the highest AUC values (Yi et al., 2023). The integration of metabolomic analysis with machine learning algorithms is evidently crucial in CRC diagnosis. Furthermore, using Explainable AI (XAI) for CRC classification based on gut microbiome data, more personalized CRC biomarker identification is performed using the Shapley Additive Explanations (SHAP) method. For instance, in the analysis of five independent datasets, the SHAP method is an effective tool for separating CRC patients into different CRC probability groups and identifying different bacteriome biomarkers in these groups (Rynazal et al., 2023). Models using multiple taxonomic biomarkers have been reported to perform better than those using single taxonomic biomarkers (AUC: 0.83). Additionally, bacterial-fungal coabundance analyses have shown interactions, particularly between \u003cem\u003eTalaromyces islandicus\u003c/em\u003e and \u003cem\u003eClostridium saccharobutylicum\u003c/em\u003e bacteria (Liu et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eFinally, in a study using mRNA meta-sequencing analysis to determine differences between the gut microbiomes of healthy individuals and CRC patients, the developed model was validated using independent test datasets and detected CRC from Stage I to Stage II with 60% accuracy. Additionally, no statistically significant differences were observed in true positive rates among samples collected from different cities or different regions of the colorectum (Konishi et al., 2022). These comprehensive findings demonstrate that the Random Forest algorithm and other machine learning techniques are powerful and effective tools in CRC diagnosis and biomarker identification. Furthermore, microbiome analyses play a significant role in this process, providing valuable information for clinical applications. In this context, integrating machine learning (ML) with metagenomic data has opened new pathways to improve diagnostic efficiency and understand the role of the microbiome in CRC. The findings of this study, consistent with previous research, underscore the transformative potential of these approaches. ML and microbiome analyses offer significant advantages not only in CRC diagnosis but also in personalized medicine applications. However, our study has some limitations. The small sample size limits the generalizability of the findings, necessitating larger cohort studies. The collection of metagenomic and tissue-specific data from CRC patients is limited due to ethical regulations and hospital procedures. The process of obtaining biological samples and patient data requires voluntary participation, making data acquisition time-consuming. High-resolution metagenomic analyses are financially expensive, posing challenges for large-scale sampling. Additionally, the computational power required for data processing can restrict working with large datasets. Small datasets can be sufficient for testing specific hypotheses and allow for more detailed biological interpretations. Preliminary studies assess accuracy and biological relevance before transitioning to larger datasets, providing scientific justification for limited sample usage. Most previous CRC studies have utilized similar dataset sizes, and our study remains consistent with the sample sizes used in current biomedical research. Findings from small-scale datasets should be validated through larger studies, and we plan to test our results on larger populations in future phases. The lack of diversity within the study cohort reduces the applicability of the results to broader populations. The data were collected from a single center, and multicenter studies are required to validate the findings. The gut microbiome is influenced by various uncontrollable factors. Future studies should ensure comprehensive ethical review and informed consent. The exclusion of multi-omic data is a significant limitation. Addressing these limitations will help improve ML models and enhance their clinical applicability in CRC diagnosis.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study underscores the transformative potential of machine learning (ML) when integrated with metagenomic and tissue-specific data to enhance colorectal cancer (CRC) diagnostics. The inclusion of demographic data further amplifies the predictive capabilities of these models, emphasizing the value of a holistic approach in medical diagnostics. The findings of this study reveal that critical bacterial biomarkers, identified through meticulous feature selection, play a pivotal role in distinguishing CRC patients from healthy controls. These results resonate with previous research that has shown improved diagnostic efficiency of CRC subgroups through ML models based on intestinal microbes. The iterative process of optimizing feature selection by removing less informative bacterial data not only enhances model performance but also sharpens our understanding of the microbiome's intricate role in CRC. This approach provides a pathway to more precise, non-invasive diagnostic tools, potentially revolutionizing CRC detection and monitoring. Future research should aim to validate these findings across larger and more diverse cohorts to ensure broader applicability. Additionally, integrating other data layers such as metabolomic and transcriptomic profiles could further refine these models, paving the way for even greater diagnostic precision and personalized treatment strategies. The application of ML to metagenomic data represents a significant leap forward in CRC diagnostics, offering new insights into the microbiome's role and setting the stage for advanced, data-driven medical solutions. This study not only contributes to the growing body of knowledge but also lays the groundwork for future innovations in cancer diagnosis and treatment.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe are grateful to the Balcalı hospital institution and the patients for their contributions to this metagenomic study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthorship Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA.D. Y.\u0026Uuml;. F.A. U.O. U.U. E.G. S.D. All authors were involved in study design, analysis and interpretation data. A.D. study conception and design, analysis and interpretation of data, and manuscript outline. F.A. U.O. U.U. E.G. analysis and interpretation of data and manuscript outline. A.D., Y.U. obtaining data and critical revision of the manuscript, analysis and interpretation of data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by the Turkish Gastroenterology Association.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAvailability of data and materials Genomic sequences used in this study have been uploaded to the NCBI GenBank database. The relevant datasets are publicly available under the following accession numbers (Sequence Read Archive submission: SUB14430300, BioProject ID: PRJNA1112781)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study complied with the ethical guidelines of the Declaration of Helsinki. Approval for this study was received by the \u0026Ccedil;ukurova University Ethics Committee on November 3, 2023 (institutional review board and clinical trial number, IRB numbers: 2023-26-138). Informed consent was obtained from all participants. All authors maintained relevant ethical standards to preserve research integrity and ensured that no duplication, falsification, or plagiarism occurred.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAbdul Rahman, H., Ottom, M. A., \u0026amp; Dinov, I. D. (2023). Machine learning-based colorectal cancer prediction using global dietary data. BMC cancer, 23(1), 144.\u003c/li\u003e\n\u003cli\u003eAli, Z. M., Hassoon, N. H., Ahmed, W. S., \u0026amp; Abed, H. N. (2020). The application of data mining for predicting academic performance using k-means clustering and na\u0026iuml;ve bayes classification. International Journal of Psychosocial Rehabilitation, 24(03), 2143-2151.\u003c/li\u003e\n\u003cli\u003eAnderson, S. M., \u0026amp; Sears, C. L. (2023). Role of the Gut Microbiome in Cancer: A Review, With Special Focus on Colorectal Neoplasia and Clostridioides difficile. Clinical Infectious Diseases, 77(Supplement 6), S471-S478.\u003c/li\u003e\n\u003cli\u003eArnold, M., Sierra, M. S., Laversanne, M., Soerjomataram, I., Jemal, A., \u0026amp; Bray, F. (2017). Global patterns and trends in colorectal cancer incidence and mortality. Gut, 66(4), 683-691. \u003c/li\u003e\n\u003cli\u003eBaba, Y., Huttenhower, C., Nosho, K., Tanaka, N., Shima, K., Hazra, A., ... \u0026amp; Ogino, S. (2010). Epigenomic diversity of colorectal cancer indicated by LINE-1 methylation in a database of 869 tumors. Molecular cancer, 9, 1-17.\u003c/li\u003e\n\u003cli\u003eBoser, B. E., Guyon, I. M., \u0026amp; Vapnik, V. N. (1992, July). A training algorithm for optimal margin classifiers. In Proceedings of the fifth annual workshop on Computational learning theory (pp. 144-152).\u003c/li\u003e\n\u003cli\u003eBray, J. R., \u0026amp; Curtis, J. T. (1957). An Ordination of the Upland Forest Communities of Southern Wisconsin. \u003cem\u003eEcological Monographs\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(4). https://doi.org/10.2307/1942268\u003c/li\u003e\n\u003cli\u003eChen, S., Webb, G. I., Liu, L., \u0026amp; Ma, X. (2020). A novel selective na\u0026iuml;ve Bayes algorithm. Knowledge-Based Systems, 192, 105361.\u003c/li\u003e\n\u003cli\u003eCortes, C. (1995). Support-Vector Networks. Machine Learning.\u003c/li\u003e\n\u003cli\u003eDekker, E., Tanis, P. J., Vleugels, J. L., Kasi, P. M., \u0026amp; Wallace, M. B. (2019). Colorectal cancer. The Lancet, 394(10207), 1467-1480.\u003c/li\u003e\n\u003cli\u003eDelik, A., Din\u0026ccedil;er, S., \u0026Uuml;lger, Y., Akkız, H., \u0026amp; Karaoğullarından, \u0026Uuml;. (2022). Metagenomic identification of gut microbiota distribution on the colonic mucosal biopsy samples in patients with non-alcoholic fatty liver disease. \u003cem\u003eGene\u003c/em\u003e, \u003cem\u003e833\u003c/em\u003e. https://doi.org/10.1016/j.gene.2022.146587\u003c/li\u003e\n\u003cli\u003eFreitas, P., Silva, F., Sousa, J. V., Ferreira, R. M., Figueiredo, C., Pereira, T., \u0026amp; Oliveira, H. P. (2023). Machine learning-based approaches for cancer prediction using microbiome data. \u003cem\u003eScientific Reports\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(1), 11821.\u003c/li\u003e\n\u003cli\u003eGuyon, I., Boser, B., \u0026amp; Vapnik, V. (1992). Automatic capacity tuning of very large VC-dimension classifiers. Advances in neural information processing systems, 5.\u003c/li\u003e\n\u003cli\u003eHarris, C. R., Millman, K. J., van der Walt, S. J., Gommers, R., Virtanen, P., Cournapeau, D., Wieser, E., Taylor, J., Berg, S., Smith, N. J., Kern, R., Picus, M., Hoyer, S., van Kerkwijk, M. H., Brett, M., Haldane, A., del R\u0026iacute;o, J. F., Wiebe, M., Peterson, P., \u0026hellip; Oliphant, T. E. (2020). Array programming with NumPy. In \u003cem\u003eNature\u003c/em\u003e (Vol. 585, Issue 7825). https://doi.org/10.1038/s41586-020-2649-2\u003c/li\u003e\n\u003cli\u003eHunter, J. D. (2007). Matplotlib: A 2D graphics environment. \u003cem\u003eComputing in Science and Engineering\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(3). https://doi.org/10.1109/MCSE.2007.55\u003c/li\u003e\n\u003cli\u003eJones, J., Shi, Q., Nath, R. R., \u0026amp; Brito, I. L. (2024). Keystone pathobionts associated with colorectal cancer promote oncogenic reprograming. Plos one, 19(2), e0297897.\u003c/li\u003e\n\u003cli\u003eKho, Z. Y., \u0026amp; Lal, S. K. (2018). The human gut microbiome \u0026ndash; A potential controller of wellness and disease. Frontiers in Microbiology, 9, 1835.\u003c/li\u003e\n\u003cli\u003eLiu, G., Su, L., Kong, C., Huang, L., Zhu, X., Zhang, X., ... \u0026amp; Wang, J. (2024). Improved diagnostic efficiency of CRC subgroups revealed using machine learning based on intestinal microbes. \u003cem\u003eBMC gastroenterology\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(1), 315.\u003c/li\u003e\n\u003cli\u003eMcKinney, W., \u0026amp; Team, P. D. (2015). Pandas-Powerful python data analysis toolkit. \u003cem\u003ePandas\u0026mdash;Powerful Python Data Analysis Toolkit\u003c/em\u003e, \u003cem\u003e1625\u003c/em\u003e. https://doi.org/10.5281/zenodo.3509134\u003c/li\u003e\n\u003cli\u003eNeuwirth, E. (2014). RColorBrewer: ColorBrewer palettes. \u003cem\u003eR Package Version 1.1-2\u003c/em\u003e, https://cran.R-project.org/package=RColorBrewer. https://doi.org/10.32614/CRAN.package.RColorBrewer\u003c/li\u003e\n\u003cli\u003eOh, J. H., Lee, J. H., Cho, M. S., Kim, H., Chun, J., Lee, J. H., Yoon, Y., \u0026amp; Kang, W. (2021). Characterization of gut microbiome in Korean patients with metabolic associated fatty liver disease. \u003cem\u003eNutrients\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(3). https://doi.org/10.3390/nu13031013\u003c/li\u003e\n\u003cli\u003eOhata, E. F., Chagas, J. V. S. D., Bezerra, G. M., Hassan, M. M., de Albuquerque, V. H. C., \u0026amp; Filho, P. P. R. (2021). A novel transfer learning approach for the classification of histological images of colorectal cancer. The Journal of Supercomputing, 1-26.\u003c/li\u003e\n\u003cli\u003ePedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M., \u0026amp; Duchesnay, \u0026Eacute;. (2011). Scikit-learn: Machine learning in Python. \u003cem\u003eJournal of Machine Learning Research\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e. https://www.jmlr.org/papers/volume12/pedregosa11a/pedregosa11a.pdf?ref=https:/\u003c/li\u003e\n\u003cli\u003ePerez, J. G., \u0026amp; Perez, E. S. (2021). Predicting student program completion using Na\u0026iuml;ve Bayes classification algorithm. International Journal of Modern Education and Computer Science, 12(3), 57.\u003c/li\u003e\n\u003cli\u003eRebersek, M. (2021). Gut microbiome and its role in colorectal cancer. \u003cem\u003eBMC cancer\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(1), 1325.\u003c/li\u003e\n\u003cli\u003eRooks, M. G., \u0026amp; Garrett, W. S. (2016). Gut microbiota, metabolites, and host immunity. Nature Reviews Immunology, 16(6), 341-352.\u003c/li\u003e\n\u003cli\u003eSameni, F., Elkhichi, P. A., Dadashi, A., Sadeghi, M., Goudarzi, M., Eshkalak, M. P., \u0026amp; Dadashi, M. (2025). Global prevalence of Fusobacterium nucleatum and Bacteroides fragilis in patients with colorectal cancer: an overview of case reports/case series and meta-analysis of prevalence studies. BMC gastroenterology, 25(1), 71.\u003c/li\u003e\n\u003cli\u003eSch\u0026ouml;lkop, B., Burgest, C., \u0026amp; Vapnik, V. (1995, August). Extracting support data for a given task. In Proceedings, First International Conference on Knowledge Discovery \u0026amp; Data Mining. AAAI Press, Menlo Park, CA (pp. 252-257).\u003c/li\u003e\n\u003cli\u003eSch\u0026ouml;lkopf, B., Burgest, C., \u0026amp; Vapnik, V. (1996). Incorporating invariances in support vector learning machines. In Artificial Neural Networks\u0026mdash;ICANN 96: 1996 International Conference Bochum, Germany, July 16\u0026ndash;19, 1996 Proceedings 6 (pp. 47-52). Springer Berlin Heidelberg.\u003c/li\u003e\n\u003cli\u003eShafi, A. S. M., Molla, M. I., Jui, J. J., \u0026amp; Rahman, M. M. (2020). Detection of colon cancer based on microarray dataset using machine learning as a feature selection and classification techniques. SN Applied Sciences, 2, 1-8.\u003c/li\u003e\n\u003cli\u003eShafi, M. A., \u0026amp; Rusiman, M. S. (2015). The use of fuzzy linear regression models for tumor size in colorectal cancer in hospital of Malaysia. Applied Mathematical Sciences, 9(56), 2749-2759.\u003c/li\u003e\n\u003cli\u003eShariati, A., Razavi, S., Ghaznavi-Rad, E., Jahanbin, B., Akbari, A., Norzaee, S., \u0026amp; Darban-Sarokhalil, D. (2021). Association between colorectal cancer and Fusobacterium nucleatum and Bacteroides fragilis bacteria in Iranian patients: a preliminary study. Infectious agents and cancer, 16(1), 41.\u003c/li\u003e\n\u003cli\u003eSiegel, R. L., Miller, K. D., \u0026amp; Jemal, A. (2020). Cancer statistics, 2020. CA: A Cancer Journal for Clinicians, 70(1), 7-30.\u003c/li\u003e\n\u003cli\u003eSpellerberg, I. F., \u0026amp; Fedor, P. J. (2003). A tribute to Claude-Shannon (1916-2001) and a plea for more rigorous use of species richness, species diversity and the \u0026ldquo;Shannon-Wiener\u0026rdquo; Index. \u003cem\u003eGlobal Ecology and Biogeography\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(3). https://doi.org/10.1046/j.1466-822X.2003.00015.x\u003c/li\u003e\n\u003cli\u003eTharwat, M., Sakr, N. A., El-Sappagh, S., Soliman, H., Kwak, K. S., \u0026amp; Elmogy, M. (2022). Colon cancer diagnosis based on machine learning and deep learning: modalities and analysis techniques. Sensors, 22(23), 9250.\u003c/li\u003e\n\u003cli\u003eTripathi, A., Misra, A., Kumar, K., \u0026amp; Chaurasia, B. K. (2023). Optimized Machine Learning for classifying colorectal tissues. SN Computer Science, 4(5), 461.\u003c/li\u003e\n\u003cli\u003eVapnik, V. N. (1963). Pattern recognition using generalized portrait method. Automation and remote control, 24(6), 774-780.\u003c/li\u003e\n\u003cli\u003eVapnik, V., Golowich, S., \u0026amp; Smola, A. (1996). Support vector method for function approximation, regression estimation and signal processing. Advances in neural information processing systems, 9.\u003c/li\u003e\n\u003cli\u003eVirtanen, P., Gommers, R., Oliphant, T. E., Haberland, M., Reddy, T., Cournapeau, D., Burovski, E., Peterson, P., Weckesser, W., Bright, J., van der Walt, S. J., Brett, M., Wilson, J., Millman, K. J., Mayorov, N., Nelson, A. R. J., Jones, E., Kern, R., Larson, E., \u0026hellip; V\u0026aacute;zquez-Baeza, Y. (2020). SciPy 1.0: fundamental algorithms for scientific computing in Python. \u003cem\u003eNature Methods\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(3). https://doi.org/10.1038/s41592-019-0686-2\u003c/li\u003e\n\u003cli\u003eWang, X., Sun, X., Chu, J., Sun, W., Yan, S., \u0026amp; Wang, Y. (2023). Gut microbiota and microbiota-derived metabolites in colorectal cancer: enemy or friend. \u003cem\u003eWorld Journal of Microbiology and Biotechnology\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(11), 291.\u003c/li\u003e\n\u003cli\u003eWickham, H., Averick, M., Bryan, J., Chang, W., McGowan, L., Fran\u0026ccedil;ois, R., Grolemund, G., Hayes, A., Henry, L., Hester, J., Kuhn, M., Pedersen, T., Miller, E., Bache, S., M\u0026uuml;ller, K., Ooms, J., Robinson, D., Seidel, D., Spinu, V., \u0026hellip; Yutani, H. (2019). Welcome to the Tidyverse. \u003cem\u003eJournal of Open Source Software\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e(43).https://doi.org/10.21105/joss.01686\u003c/li\u003e\n\u003cli\u003eZeng, R., Gou, H., Lau, H. C. H., \u0026amp; Yu, J. (2024). Stomach microbiota in gastric cancer development and clinical implications. Gut, 73(12), 2062.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-applied-genetics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"joag","sideBox":"Learn more about [Journal of Applied Genetics](https://www.springer.com/journal/13353)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/joag/default.aspx","title":"Journal of Applied Genetics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"colorectal cancer, metagenomic, machine learning, diagnosis, prognosis","lastPublishedDoi":"10.21203/rs.3.rs-7808541/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7808541/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eColorectal cancer (CRC) represents a significant global health burden. Leveraging machine learning (ML) with metagenomic and tissue-specific data presents new opportunities for improving diagnostic accuracy and understanding the microbiome's role in CRC.\u003c/p\u003e\u003ch2\u003eObjective\u003c/h2\u003e\u003cp\u003eThis study aimed to enhance diagnostic efficiency and identify crucial bacterial biomarkers in CRC using various ML models applied to metagenomic data.\u003c/p\u003e\u003ch2\u003eMaterial and Methods\u003c/h2\u003e\u003cp\u003eA total of 33 samples were analyzed, comprising 20 healthy controls and 13 CRC patients. Each sample included demographic data (age, gender) and bacterial information (\u003cem\u003eBacteriodes\u003c/em\u003e, \u003cem\u003eEnterococcus\u003c/em\u003e, \u003cem\u003eFaecalibacterium\u003c/em\u003e, \u003cem\u003eProteobacteria\u003c/em\u003e, \u003cem\u003eGammaproteobacteria\u003c/em\u003e, \u003cem\u003eFirmicutes\u003c/em\u003e, \u003cem\u003eEnterobacteriaceae\u003c/em\u003e, \u003cem\u003eClostridia\u003c/em\u003e). We employed five ML models: Linear Regression, Naive Bayes, Multilayer Perceptron (MLP), Support Vector Machine (SVM), and Decision Tree. Using leave-one-out cross-validation, we evaluated model accuracy and optimized feature selection by iteratively removing bacterial data to improve model performance.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eThe highest accuracy was 87.88% in the presence of demographic information, while the highest accuracy was 84.85% in the absence of demographic information. In the presence of demographic information, the highest accuracy was achieved with the MLP model, and in the absence of demographic information, with the SVM linear model. While the MLP model required 5 bacterial information to reach a accuracy of 87.88%, the SVM linear model could reach a accuracy of 84.85% with 6 bacterial information.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e\u003cp\u003eMachine learning models, particularly MLP, combined with metagenomic data, can significantly improve CRC diagnostic accuracy. Identifying key bacterial biomarkers enhances our understanding of the microbiome's role in CRC, providing a foundation for non-invasive diagnostic tools and personalized treatment strategies.\u003c/p\u003e","manuscriptTitle":"Machine Learning Integration of Tissue-Specific Metagenomic Signatures for Colorectal Cancer Diagnosis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-02 13:40:52","doi":"10.21203/rs.3.rs-7808541/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major Revisions Needed","date":"2026-01-21T02:08:37+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2025-11-27T09:44:06+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-25T11:21:26+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-09T02:41:49+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Applied Genetics","date":"2025-10-08T09:47:59+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-applied-genetics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"joag","sideBox":"Learn more about [Journal of Applied Genetics](https://www.springer.com/journal/13353)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/joag/default.aspx","title":"Journal of Applied Genetics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"66f10001-b672-414f-98e3-e0aaed6003bb","owner":[],"postedDate":"December 2nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-15T15:17:39+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-02 13:40:52","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7808541","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7808541","identity":"rs-7808541","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}