Meta-analysis of Multi-functional Biomarkers for Discovery and Predictive Modeling of Colorectal Adenoma and Carcinoma

Meta-analysis of Multi-functional Biomarkers for Discovery and Predictive Modeling of Colorectal Adenoma and Carcinoma | 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 Method Article Meta-analysis of Multi-functional Biomarkers for Discovery and Predictive Modeling of Colorectal Adenoma and Carcinoma Scott N. Peterson, Alexey M. Eroshkin, Piotr Z. Kozbial, Ermanno Florio, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2838129/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: Despite the effectiveness of colonoscopy for reducing colorectal cancer (CRC) mortality, poor screening compliance ranks CRC as the second most deadly malignancy. There is a need to develop a preventative, non-invasive diagnostic test, such as a fecal microbiota test, for early detection of both pre-cancerous adenomas and carcinomas to effectively reduce mortality. Results: We conducted a clinical meta-analysis of published deep metagenomic stool sequence datasets including 1,670 subjects from 9 countries, including 703 healthy controls, 161 precancerous colorectal adenoma (CRA), 48 advanced precancerous colorectal adenoma (CRAA) and 758 CRC cases diagnosed by colonoscopy. We analyzed these data through a novel automated machine learning workflow using a two-stage feature importance ranking and ensemble modeling method to identify and select highly predictive taxonomic and functional biomarkers. Machine learning modeling of selected features differentiated the metagenomic profiles of healthy patients from CRA, CRAA and CRC cases with an average area under the curve (AUC) for external holdout testing of 0.84 (sensitivity=0.82; specificity=0.71, accuracy=0.77) for CRC; an AUC of 0.97 (sensitivity=0.78; specificity=0.98, accuracy=0.97) for CRAA; and an AUC of 0.90 (sensitivity=0.74, specificity=0.89, accuracy=0.86) for CRA. These performance outcomes represented a 2%, 3% and 8% increase in AUC, compared to baseline ML performance, respectively. The predictive features identified for each disease class were largely distinct and represented differing proportions of taxonomic and functional features. Conclusions: The predictive taxonomic features identified for each disease class were largely distinct, whereas many functional gene features were shared across disease classes but displayed differing direction of change. Application of our ensemble approach for feature selection increased the predictive power of each disease class and moreover may generate discriminatory models with greater generalizability. colorectal cancer stool microbiome artificial intelligence machine learning biomarkers Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Background Colorectal cancers are among the most prevalent cancers world-wide with an estimated 1.8 million new colon cancer cases and over 700,000 rectal cancer cases reported in 2018 [ 1 ]. Despite the strong evidence demonstrating that screening of individuals with average CRC risk reduces mortality [ 2 ], compliance amongst individuals is limited due to the invasiveness, discomfort and fear associated with colonoscopy. This has created a significant gap in CRC prevention within the health care system, emphasizing the need for sensitive, accurate, non-invasive diagnostic tests to detect colonic adenomas and carcinomas. It is expected that non-invasive tests will increase screening compliance over time. Our efforts seek to fill this gap by developing a powerful set of artificial intelligence (AI) driven methods to identify diagnostic biomarkers derived from microbial species present in stool samples. Among the non-invasive CRC detection tests is the Fecal Immunochemical Test (FIT) that is associated with limited sensitivity (79%) for detecting CRC [ 3 ] and a poor sensitivity (~ 25%) for detecting advanced adenomas [ 4 ]. A multi-target stool assay (Cologuard), measures KRAS mutations, aberrant NDRG4 and BMP3 methylation together with b-actin and hemoglobin immunoassays, performs better than FIT, detecting CRC cases with greater sensitivity (AUC = 92% compared to 74% for FIT alone), however advanced premalignant lesions were still poorly detected by the Cologuard and FIT test assays (~ 42% and ~ 24% respectively) [ 5 ]. These outcomes highlight another important gap in the healthcare system based on the relatively poor ability of existing non-invasive methods to detect early and advanced adenomas. Development of a diagnostic that addresses this gap will be of high impact both in terms of improved detection of pre-malignant lesions and potentially an overall reduction of colonoscopies required for average risk subjects. CRC is a heterogeneous disease, the majority of cases are sporadic without underlying heritable features [ 6 ]. A wide variety of environmental factors including a western diet, obesity, cigarette smoking, alcohol consumption and lack of exercise are known CRC risk factors [ 7 – 9 ]. Chief amongst these risk factors is diet, that was linked to an estimated ~ 38% of incipient CRC cases. Additional evidence for environmental influence of CRC is based on findings that the incidence of CRC is influenced by emigration, wherein a subject’s risk of CRC development is altered based on the diet and lifestyle of the recipient country [ 10 ]. Each of the above-mentioned CRC risk modifiers is also known to modulate the composition of the gut microbiota [ 11 – 14 ]. This association has drawn substantial attention to the gut microbiota as a potential mediator of CRC initiation and/or progression, and accumulating evidence supports this hypothesis. The large number of species and genes encoded in the gut microbiome represents an attractive source of potential biomarkers for diagnostics and prognostics of early, premalignant adenomas, advanced adenomas, and CRC. Several studies have examined gut microbiota in CRC using either 16S rRNA or shotgun metagenomic sequencing. These studies have explored fecal and mucosal-associated microbial populations and different stages along the adenoma, carcinoma progression. One previous meta-analysis of fecal microbiota datasets resulted in the identification of seven bacterial species enriched in CRC ( Bacteroides fragilis, Fusobacterium nucleatum, Parvimonas micra, Porphyromonas assacharolytica, Prevotella intermedia, Alistipes finegoldii and Thermoanaeroovibrio acidaminovorans [ 15 ]. A separate pair of meta-analyses identified an expanded set of twenty-nine species enriched over eight distinct geographical regions [ 16 , 17 ]. A number of studies have analyzed the human gut microbiota associated with colonic tumors and normal adjacent tissue, leading to the identification of dysbiotic signatures associated with CRC. While specific taxa vary from study to study, some common themes have emerged including the frequent identification of elevated relative abundance of E. coli [ 18 ], Fusobacterium nucleatum [ 19 ] and enterotoxin-producing Bacteroides fragilis (ETBF) strain [ 20 ]. Additional taxa associated with CRC have also been identified, but are less uniformly observed across studies. An important observation established by these studies is that the magnitude of differentiation in relative abundance of microbial biomarkers derived from tissue samples is substantially greater than that of stool samples. In stool samples, differentially abundant taxa are more subtle, often requiring AI-based methods for detection. A number of characteristics of fecal microbiota present specific challenges in the identification of diagnostic biomarkers for early detection of adenomas and carcinomas, including high dimensionality, data sparsity, and low generalizability. Despite the massive quantity of DNA sequence data generated in shotgun metagenomic sequencing of stool samples, we have observed that the best performing taxonomic biomarkers are detected in only a relatively small number of samples. This exemplifies the problem of data sparsity and dictates that a high-performance diagnostic test based on next generation sequencing (NGS) sequence data will require multiple independent biomarkers to compensate for low prevalence of any single microbial biomarker in the human population. Here we describe a series of methodological improvements to a diagnostic NGS data analysis pipeline that addresses these challenges. Our pipeline automates several steps that increase efficiency and performance of available models, ultimately improving the power of taxonomic and KEGG ortholog (KO), gene features to distinguish healthy subjects from those with early and advanced adenomas and those with carcinomas. An overarching goal of this study was to improve the generalizability of machine-learning models that avoid a variety of known pitfalls associated with metagenomic data, These pitfalls include the use of heterogeneous methods and analysis procedures for data generation. Our analysis of metagenomic datasets indicate distinctions in the most informative biomarkers for each disease class. Features of highest importance distinguishing CRC from healthy controls were disproportionately reliant on taxonomic features, whereas CRA and CRAA features were more balanced in representation of KO and taxonomic features. We note that the optimal features for each disease class display very little overlap, suggesting that the adenoma-to-carcinoma progression involves uniquely selective biochemical environments for fecal microbiota that do not follow a simple linear relationship. These results are discussed with an outlook toward future studies, that may facilitate increased diagnostic power, deciphered from fecal microbiota. Methods Data processing. The raw fecal shotgun DNA sequence data was preprocessed and taxonomically profiled using the bioBakery 3 pipeline v3.0.0-alpha.7 using the default parameters and kneaddata v0.10.0, MetaPhlAn version 3.0.7, HUMAnN v3.0.0.alpha.4, biobakery_workflows. Functional profiling of metagenomic sequencing data was performed using HUMAnN 3 genefamilies.tsv output and KEGG orthology gene groups (KO). To facilitate data interpretation and visualization, we regrouped gene family abundance data (represented by UniRef identifiers) into KO functional categories using humann_regroup_table script. Data normalization and visualization by PCoA. Total sequence read counts including taxa from all ranks (kingdom to species) were normalized using weighted trimmed mean of M-values (TMM) using the ‘edgeR’ package[ 21 ] and converted into log-counts per million (log-CPM) using the voom function implemented in the ‘limma’ package [ 22 ] in R version 4.2.1. The data were further normalized using a supervised normalization method (SNM) as described [ 23 ] to remove significant batch effects between projects while retaining biological differences between disease classes. The SNM method was implemented in the ‘snm’ package [ 24 ] in R. The effects of supervised normalization were visualized using Principal Coordinate Analysis (PCoA). PCoA was performed using Euclidean distances on both TMM-Voom and TMM-Voom-SNM transformed count tables. Differences in the microbial composition between projects and disease class were assessed separately using the adonis2 function from the ‘vegan’ package ( https://CRAN.R-project.org/package=vegan ). Mean distance of samples from the centroids in the PCoA plots was compared between projects using the Kruskal-Wallis test. Machine learning. Models were created using the automated machine learning platform called DataRobot (DR); ( www.datarobot.com ). A custom python script was developed for automated data submission to DR that allows the development of models for multiple datasets. The best model of all developed models was selected based on the largest area under the curve (AUC) value for external test dataset prediction. “Blender models” which are obtained using several machine learning algorithms (combining the predictions of two or more models), were not used here. For classification purpose for each disease target, the set of samples was divided randomly into a training set (80% of samples) and a test set (20% of samples). The training set was used to develop a set of high performing predictive models in DR (using more than ten different machine learning architectures for classification such as eXtreme Gradient Boosted Trees Classifier, Keras Slim Residual Neural Network Classifier using Training Schedule, Elastic-Net Classifier and Light Gradient Boosted Trees Classifier with Early Stopping). The developed models were used to predict the disease state in the remaining 20% of samples (unseen data which were not used in training), and the top model (with highest external test AUC) is defined. The model performance on the test set parameters was then supplemented with external test sensitivity, specificity, and accuracy. DataRobot feature lists. Feature lists control the subset of features that DataRobot used to build models. DataRobot automatically creates several feature lists for each project, including Informative Features and DR Reduced Features. Informative Features are all features that provide information potentially valuable for modeling (normally all features). DR Reduced Features are a subset of features, selected based on the Feature Impact calculation of the best model. The DR Reduced feature list consisted of the features that provide 95% of the accumulated impact for the model. Since Informative Feature lists usually have almost all features in the dataset (1,000–2,000 in taxonomy annotation and 5,000–10,000 in functional annotation) and DR Reduced feature list have no more than 100 features, for comparative purposes we used models built using DR Reduced feature lists. Permutation-based Feature Impact measures a drop in model accuracy when a feature’s values are shuffled. To compute these values, DataRobot makes predictions on a sample of training records and then alters the training data (shuffles value) computing a drop in accuracy that resulted from shuffling. This shuffling process was repeated and the results were normalized and ranked (top feature has an impact of 100%). The sampling process corresponds to one of the following criteria: For balanced data, random sampling is used. For imbalanced binary data, smart down sampling is used; DataRobot attempts to make the distribution for imbalanced binary targets closer to 1:1 and adjusts the sample weights used for scoring. For zero-inflated regression data, smart down sampling was used; DataRobot groups the non-zero elements into the minority class. For imbalanced multiclass data, random sampling was used. FIRE feature selection. A feature reduction and selection method “Feature Importance Rank Ensembling” (FIRE, https://www.datarobot.com/blog/using-feature-importance-rank-ensembling-fire-for-advanced-feature-selection/ ) was used. In this method, the features are derived from multiple diverse predictive models built by DR. These models were sorted by external test AUC. The median rank of each feature was calculated by aggregating the ranks for each of the top models (the number of top models to consider was empirically selected equal to five). The FIRE procedure consists of the following steps: (a) calculate the feature importance for the top 5 models (determined by the external test AUC), (b) obtain the ranking of the features, (c) compute and sort the aggregated list by the computed median rank, (d) define the threshold number of features to select, (e) define a feature list based on the newly selected features. Since the optimal number of features is not known, we iteratively tested several thresholds with large increments in the first loop (800, 700, 600, 500, 400, 300, 200, 100, 90, 80, 70, 60, 50, 40, 30) and small increments around the first found threshold in the second loop (for example, 85, 84, 83, 82, 81, 79, 78, 77, 76, 75, if the best threshold from the first loop equals 80). We took the threshold that provided the highest external test AUC. We considered the maximal number of FIRE features as 800 since in some of the analyzed datasets, the total feature number was between 800 and 900. We implemented FIRE in a multi-step process to exploit ensemble approaches to establish feature selection with increased generalizability since its basis is not tied to a single model and its associated biases. SIAMCAT feature selection. Another feature selection method used here is based on statistical inference of associations between microbial communities and phenotypes and is referred to as SIAMCAT version 2.1.0 [ 25 ]. SIAMCAT is part of the suite of microbiome analysis tools developed at EMBL. We applied the cut-off of adjusted p-value < 0.001) to identify important features. Combining FIRE and SIAMCAT. Since the FIRE and SIAMCAT feature lists are selected based on different criteria (ensemble ranking for the first and statistical significance for the second), we also tested feature lists that combine the FIRE and SIAMCAT features together to create new feature list. Results To evaluate the accuracy and feasibility of developing a non-invasive diagnostic stool test for early and advanced pre-cancerous adenomas and carcinomas, we imported data generated from 11 studies conducted by laboratories in 9 countries, analyzing stool samples by shotgun metagenomic sequencing [ 16 , 17 , 26 – 34 ]. All samples were confirmed by colonoscopy. The descriptive statistics of the cohorts from each study are shown in Table 1 . Table 1 Selected Metagenomic projects used for modeling. Project Number Gender Age BMI Country disease type raw reads post-qc reads (%) human reads Spanogiannopoulis (34) 10 Male:5 female 5 52 ± 20.05 Missing USA CTR: 0 CRC: 10 CRA: 0 CRAA: 0 37213448 ± 6907944 34520932 ± 6316395 4.00 ± 9.49 Feng (27) 156 male: 88 female: 68 66.9 ± 8.32 27.4 ± 4.02 Austria CTR: 63 CRC: 46 CRA: 0 CRAA: 47 52689474 ± 8343659 46088635 ± 7292627 4.63 ± 1.05 Gao (30) 126 Missing Missing Missing China CTR: 47 CRC: 39 CRA: 40 CRAA: 0 46462323 ± 16612805 42959333 ± 15584240 1.45 ± 3.55 Gupta (31) 30 male: 18 female: 11 missing: \ 1 59.8 ± 7.81 Missing India CTR: 0 CRC: 30 CRA: 0 CRAA: 0 9229167 ± 4142109 8510190 ± 3816170 1.7 ± 0.651 Hannigan (35) 81 male: 46 female: 35 58.6 ± 10.8 28.1 ± 6.1 missing: 1 Canada, USA CTR: 28 CRC: 27 CRA: 26 CRAA: 0 6593685 ± 3784609 4964982 ± 2801436 2.69 ± 0.718 Thomas (16) 140 male: 52 female: 28 missing: 60 67.5 ± 8.73 missing: 60 25.5 ± 3.93 missing: 64 Italy CTR: 52 CRC: 61 CRA: 27 CRAA: 0 44984798 ± 24403021 41938748 ± 22955372 1.52 ± 0.501 Vogtmann (33) 104 male: 74 female: 30 61.5 ± 12.3 25.1 ± 4.25 missing: 3 USA CTR: 52 CRC: 52 CRA: 0 CRAA: 0 62406634 ± 15463669 55272649 ± 14006526 6.62 ± 2.51 Wirbel (17) 130 male: 76 female: 54 63.4 ± 12.1 24.9 ± 4.2 Germany CTR: 60 CRC: 70 CRA: 0 CRAA: 0 25277871 ± 9126431 23317638 ± 8560313 1.75 ± 0.791 Yachida (29) 611 male: 353 female: 258 61.8 ± 11 22.9 ± 3.37 missing: 10 Japan CTR: 286 CRC: 258 CRA: 67 CRAA: 0 45765841 ± 12910710 41600333 ± 11694276 1.13 ± 0.576 Yu (26) 128 male: 81 female: 47 64.2 ± 9.08 23.8 ± 3.08 missing: 1 China CTR: 54 CRC: 74 CRA: 0 CRAA: 0 56317665 ± 9956025 48374964 ± 9470724 4.35 ± 1.63 Zeller (28) 154 male: 84 female: 70 63.1 ± 12 25.5 ± 4.04 missing: 4 France, Germany CTR: 61 CRC: 91 CRA: 1 CRAA: 1 58257017 ± 23112145 50340355 ± 20826272 6.21 ± 5.54 Descriptive statistics of gender, BMI, age, disease classification, raw reads, post-qc reads, and percentage of human reads for each project. For continuous variables, mean and standard deviation are shown and for categorical variables number of samples within each category is shown. To analyze publicly available data sets we first performed data normalization as described [ 23 ] and (see methods). The effects of supervised normalization were visualized using Principal Coordinate Analysis (PCoA). Differences in the microbial composition between projects and disease class were assessed separately. PCoA plots showed that supervised normalization significantly reduced the variation that could be explained by unique projects from an R 2 = 10–0.075% (Fig. 1 A-B). Only 1.4% of the variation was explained by disease class using the TMM-Voom normalized data (Fig. 1 C). This variation decreased to 0.896% following supervised normalization, though the difference between disease types remained significant (Fig. 1 D). To identify projects and samples representing potential outliers we determined distance to centroids within each project. These distances were similar across projects for both TMM-Voom and TMM-Voom-SNM data ( Supplementary Figure S1 ). Although non-ideal, the project-specific variation was fully expected. We elected to conduct performance optimization of these datasets rather than attempting to remove studies based on ad hoc criteria. These results highlight the significant challenges associated with meta-analyses of gut microbiome data. It is difficult to distinguish between study and country-specific effects. Therefore, the possibility that biomarkers associated with adenomas and carcinomas are prone to country or regional-specific effects remains unresolved. To illustrate study-variability we used data from individual studies to train models which were then used to predict samples from all other studies to determine how well each model predicted external samples (Fig. 2 ). In most instances, study-specific models performed well on themselves. Some studies used for training generated relatively higher AUC for test sets across all or most studies. Other studies used as training sets predicted one or a few studies with high AUC but displayed greater variation overall. Finally, some studies performed relatively poorly as training sets for most or all other studies. The reasons for cross-study variability may be numerous and include sampling differences, methodological variability, read-depth, geographic effects, lesion location, cohort demographics, and others. This result may be a way to measure the generalizability of features derived from particular studies. Feature generation. In our efforts to develop a stool microbiome diagnostic analysis pipeline, we have focused on the evaluation of two related data features. The first, are the relative abundance/prevalence of taxonomic features [ 35 ]. Second, we explored the inclusion of gene features derived from shotgun metagenomic sequence analysis. The KEGG Ortholog (KO) grouping is a database of molecular functions representing functional orthologs [ 36 ]. We tested the hypothesis that KO features may positively contribute to predictive performance when combined with those derived from taxonomy. Feature Processing and Selection. We have implemented strategies to evaluate a large variety of feature reduction methods to compare their overall impact on prediction accuracy. Each of these had specific strengths and weaknesses. We expected that feature selection schemes based on filtering low-prevalence features would be risky in the context of fecal microbiota since many of the best diagnostic features (species) are of low abundance and/or low prevalence. Low prevalence dictates that the best performing models are likely to require a larger number of features for optimal accuracy. Here we evaluate the feature reduction method referred to as Feature Importance Rank Ensembling (FIRE) (see methods), in which the highest-ranking features are derived from multiple diverse models. We elected to identify the best 5 models (by an external test AUC) in our ensemble procedures. Five models were chosen empirically since a general degradation in performance was observed in the sixth best model. We added another feature selection method based on statistics referred to as SIAMCAT to define a novel workflow (Fig. 3 ). One important finding based on implementation of FIRE is that the best performing models differ according to disease class, emphasizing that no single model or set of models is optimal to distinguish health from CRA, CRAA, CRC. We therefore performed FIRE using feature selection and training on CRA, CRAA and CRC samples independently to achieve target-specific model optimization. This in turn achieves optimal disease-class-specific diagnostic performance. FIRE Feature Selection. While our results were based on a process that began with 800–1100 features, in practice, any number of features may be selected, however an optimal feature number must be determined empirically. To establish the optimal number of features for each disease class, FIRE was performed iteratively to establish a performance score based on external test AUC. We conducted these analyses for each disease target and for taxonomic and KO features separately ( Table 2 ). Table 2. FIRE Feature Selection. The table reports AUC values for the external (20%) data sets. Bold underlined values are the maximal external test AUC achieved for a particular annotation, target, and FIRE features set size. 1 eXtreme Gradient Boosted Trees Classifier, 2 Keras Slim Residual Neural Network Classifier using Training Schedule (1 Layer: 64 Units), 3 Elastic-Net Classifier (L2 / Binomial Deviance), 4 Light Gradient Boosted Trees Classifier with Early Stopping We did not observe any pattern across disease classes when evaluating taxonomic features and functional KO features separately. For CRC, 800 KO features and 400 taxonomic features provided the best performance. This was strongly contrasted by CRAA and CRA analyses. For CRAA the optimum number of KO and taxonomic features was substantially lower (40 and 100, respectively). For CRA we observed 40 KO features and 70 taxonomic features were optimal. SIAMCAT. As an additional layer of ensemble-based analysis, we processed taxonomic and gene features through SIAMCAT to allow visualization of differential abundance, prevalence, feature AUC and feature ranking based on statistical significance (Fig. 4 ). The feature importance of taxa associated with CRC (adj p < 0.001), illustrate that some taxonomic features display negligible fold-change, but display significant shifts in prevalence. This characteristic was most pronounced in CRC microbiota. Combining FIRE and SIAMCAT features. We combined features generated by FIRE and SIAMCAT. Not surprisingly, the features generated by FIRE and SIAMCAT partially overlap. The unique features from a combined list were used to classify samples into healthy or disease classes. To determine whether combining FIRE and SIAMCAT selected features improves performance we evaluated possible incremental improvements of our approach by generating AUCs using the best model from machine learning algorithms to establish a baseline for comparison to FIRE and SIAMCAT alone and in combination (Fig. 5 ). The best performance for CRA was achieved when combining taxonomic and KO features selected by combined FIRE-SIAMCAT. This approach yielded a nearly 8% increase in external AUC (baseline AUC = 0.80 vs 0.87). Analysis of CRAA performance was somewhat more complex. All feature types performed best when using FIRE alone and there was little difference in the performance when using taxonomic features alone or in combination with KO features. FIRE selected features generated a 3% gain in external AUC (baseline AUC = 0.94 vs 0.97). The results for CRC showed that the combination of taxonomic and KO features outperformed taxonomic features alone which in turn outperformed KO features alone. The best performance was observed from features selected by FIRE, resulting in a modest 2% increase in external AUC (baseline AUC = 0.80 vs 0.82). A major conclusion from this analysis is that the microbiota and microbiome associated with each disease class demand defining distinct computational workflows, as no single or set of models perform optimally on all 3 disease classes. Feature overlap across disease classes. To gain biological insights into the features that contribute most significantly to distinguishing or unifying disease class prediction, we analyzed 800 features generated from a combination of FIRE and SIAMCAT across disease classes (Fig. 6 ). Figure 6 . When comparing the top 20 features (bottom right Venn diagram), there was no overlap in either taxonomic or gene features. It is notable that 60% of the features for CRC are taxonomic, significantly larger than that observed for CRA (40%) and CRAA (25%). When examining the top 50 features (bottom middle Venn diagram), differences in the proportion of features derived from taxa or KO dissipate. Among the top 50 features we observe modest overlap in features across disease classes. Comparison of the top 100 features shows that the proportions of taxonomic features become quite even across disease classes. As more features are compared, we note that the proportion of gene features continue to increase relative to taxonomic features, and we observe increasing overlap across disease classes. Examination of 800 features reveals that among the overlaps, KO features dominate relative to taxonomic features. Indeed, we observed 48 KO features (2.4%) shared among all 3 disease classes and no cases of overlapping taxonomic features. This imbalance is also evident in all pairwise comparisons of overlapping features such that taxonomic features represent between ~ 9–14% of overlapping features. Given that shared taxonomic features frequency is similar across disease classes, it is notable that the number of shared gene features is significantly higher between CRC and CRA relative to any other pairwise relationship. To refine these comparisons, we considered the direction of change of feature types (increased or decreased in disease), to better approximate the biological similarity of shared taxonomic and gene features (Fig. 7 ). The diagram (top left) indicates that the CRA and CRC microbiome share a significantly higher number of features altered in relative abundance in the same direction compared to other pair-wise comparisons. The relationship between CRA and CRAA samples (top right) shows the small number of features displaying the same direction of change. We note that the large number of overlapping features shared between CRC vs CRAA and CRA vs CRAA) display the opposite direction of change. This result suggests that a subset of features contribute to prediction performance in all disease targets however, the direction of change of these common features in fact, distinguishes CRAA from CRA and CRC. Model validation: analysis of taxonomic features. A comparison of the features established using our ensemble feature selection approach ( Supplementary Table S1 ) to other similar large-scale meta-analyses confirmed a subset of features associated with CRC including; Fusobacterium nucleatum [ 15 – 17 , 29 ], Gemella morbillorum [ 16 , 17 , 29 ], Bifidobacterium catenulatum [ 16 ], Peptostreptococcus anaerobius [ 16 , 29 ], Peptostreptococcus stomatis [ 16 , 17 , 29 ], Porphyromonas asaccharolytica [ 15 , 16 ], Parvimonas micra [ 15 – 17 , 29 ], Solobacterium [ 16 , 17 , 29 ], Clostridium symbiosum [ 16 , 17 ], Eubacterium eligens [ 29 ], Prevotella intermedia [ 15 – 17 ], Hungatella hathewayi [ 17 ], Roseburia intestinalis [ 16 ], Fusobacterium sp oral taxon 370 [ 17 ], Bacteroides fragilis [ 16 ], Porphyromonas uenonis [ 16 , 17 , 29 ], Prevotella nigrescens [ 17 ] and Anaerococcus vaginalis [ 16 , 17 ]. These features may represent the most “universal” biomarkers for CRC that over-ride confounding country-specific effects, although other aspects of study design and technical methods and reporting are also likely to explain why more features are not conserved across studies. The best predictive KO features ( Supplementary Table S1 ) did not provide biological insight, since under- and over-represented KO’s taken from the top 800 features (KO and taxonomy) did not generate sufficient coverage of pathways to establish confidence in their potential impact on human physiology. Among the 735 KO features associated with CRA and 713 associated with CRAA, only 10 (1.4%) were altered in the same direction, whereas 122 (16.6%) were altered in opposite direction. Similarly, a comparison of the 704 KO features associated with CRC, 28 (4%) displayed the same direction of change as that observed in CRAA, whereas 108 (15.3%) were in opposite direction. Finally, a comparison of KO features in common between CRA and CRC showed that 25.4% changed in the same direction whereas only 1.7% changed in the opposite direction. We are currently conducting a comprehensive global analysis of gut microbiome functional features (in preparation) and therefore describe only taxonomic features here. In order to assess whether taxonomic features for each class represent coherent phylogenetic groups and direction of change, we analyzed important features at various phylogenetic levels. Among the top 800 features, 65 represented taxa over- or under-represented in CRA, 87 taxa for CRAA and 96 taxa for CRC. It should be noted that these features generally did not achieve statistical significance in comparisons but were deemed discriminatory based on AI models used. In total, the feature importance list contained taxa from 14 classes. The top features for all disease classes defined 41 different bacterial families. The distribution of important features for each disease class were disproportionately associated with particular phylogenetic groups (Fig. 8 and Supplementary Table S2 ). For CRAA, within the phylum, Archaea, Methanobrevibacter smithii is uniquely over-represented compared to healthy controls. Within the phylum, Actinobacteria, taxa over-represented in CRAA samples were disproportionately represented. Within the family Actinomycetaceae, A. graevenitzii, Actinomyces sp HMSC035G02, Actinomyces sp ICM47, A. viscosus were all over-represented in CRAA samples. In CRA samples A. odontolyticus was under-represented, whereas Actinomyces cardiffensis and Actinomyces turicensis were over-represented in CRC samples. Among the Bifidobacteriaceae, only 2 species ranked as informative, B. longum is over-represented in CRAA samples, whereas B. catenulatum , is under-represented in CRC samples. Among the Propionibacteriaceae, P. freudenreichii was uniquely over-represented in CRAA samples. Within the family Atopobiaceae, Atopobium rimae and Olsenella scatoligenes were both over-represented in CRAA samples. Within the family Coriobacteriaceae, Collinsella intestinalis , Collinsella stercoris and Enorma massiliensis and [ Collinsella ] massiliensis were all over-represented in CRAA samples, whereas the latter was under-represented in CRC samples. Several genera within Eggerthellaceae were over-represented in CRAA samples including; Adlercreutzia, Asaccharobacter, Enterorhabdus and Gordonibacter were over-represented in CRAA samples, whereas the latter two were under-represented in CRA samples. Slackia exigua was uniquely over-represented in CRC samples. The 11 differentially represented Bacteroides spp., were distributed across all disease classes. B. eggerthii was uniquely under-represented in CRA samples, whereas B. intestinalis, B. plebeius, B. salyersiae and B. stercoris were all over-represented. Both, B. thetaiotaomicron and B. xylanisolvens were under-represented in CRAA samples. B. fragilis and B. plebeius were over-represented in CRC. The family Porphyromonadaceae contained 3 species uniquely over-represented in CRC samples, including; P. asaccharolytica, P. endodontalis and P. uenonis . Taxonomic markers associated with CRC samples were strongly enriched for species belonging to the Prevotellaceae, including; P. intermedia, P. nigrescens and Prevotella sp CAG 520, all of which displayed increased prevalence compared to healthy controls. P. stercorea was also increased in relative abundance in CRA and CRC samples. Among the family Tannerellaceae, Parabacteroides goldsteinii and P. gordonii were both enriched in CRA, whereas P. distasonis was reduced in CRAA samples. Examination of features within Streptococcaceae, we found biomarkers associated with all three disease classes including; S. thermophilus (CRA), S. mitis (CRAA) and S. pasteurianus and S. salivarius (CRC). A large number of informative features came from the family Lachnospiraceae including; Anaerostipes hadrus , Dorea longicatena , Roseburia sp CAG 309, Roseburia sp CAG 471 all of which were reduced in CRA samples. Blautia wexlerae , Ruminococcus torques , Coprococcus catus were also increased but to a smaller extent in CRC. Coprococcus comes , Dorea formicigenerans , Dorea longicatena , Roseburia faecis were all increased in CRAA samples and Clostridium bolteae was reduced. Eisenbergiella tayi and Clostridium symbiosum were enriched in CRC samples, whereas Roseburia intestinalis and Roseburia sp CAG 303 were reduced in relative abundance. Within the family Peptostreptococcaceae, P. anaerobius and P. stomatis both displayed increased relative abundance and stronger increased prevalence in CRC samples. The family Peptoniphilaceae contained Anaerococcus vaginalis and Parvimonas micra both associated with increased relative abundance and prevalence in CRC samples. Similarly, the Fusobacteriaceae possessed three species associated with CRC, F. naviforme , F. nucleatum and Fusobacterium sp oral taxon 370. These results are indicative of biological significance since related genera and species typically behave in a coherent manner in disease. Discussion We have conducted an in-depth meta-analysis of publicly available microbiome shotgun sequence data of fecal samples derived from healthy control donors and those diagnosed by colonoscopy as CRA, CRAA and CRC. Our work describes a novel ensemble approach using two independent algorithms (FIRE and SIAMCAT) for feature selection. This process generated improved external HO AUC for each disease class compared to the best DR baseline model alone. Perhaps more important than improved accuracy of our approach is the likely benefit of increased generalizability achieved using ensemble approaches that reduce biases introduced from single model AI methods. We noted that microbiota sequence data derived from each disease class generated different unique best models ( Table 2 ). Key finding from our studies was the utility of using both taxonomic and gene features to discriminate healthy controls from those with adenomas and carcinomas. Finally, we noted that as the number of discriminatory features increased, the proportion of useful gene features increased relative to taxonomic features, suggesting that taxonomic biomarkers are more finite compared to gene features. The characteristics of the 11 studies used varied substantially, including 9 different countries, a focus on different disease states, sampling procedures, variable cohort size, and number of reads passing qc metrics (Table 1 ). Most studies attempted to balance gender and age within their respective cohorts, however male subjects in general were more prevalent than female. Additional factors such as DNA preparation and sequencing methods varied across studies, and importantly, some studies collected fecal samples after colonoscopy. All of these features are expected to create variability in study outcomes. Despite these confounding factors, the taxonomic features identified in individual studies, while variable, define a consensus finding, at least for CRC where the most high quality studies are available for meta-analysis. Given the known inter-personal variability in microbiota and distinct dietary habits of each participating country, the fact that similar taxa are identified strongly suggests that the selective forces operating in CRC are dominant to diet and other known selective pressures. We quantified the variability of the datasets used by comparing b-diversity by PCoA (Fig. 1 ) and computing Euclidean distance from centroid ( Supplementary Figure S1 ). These analyses illustrated that large and comparable sample-to-sample variability existed within each study as expected. Most revealing was the outcome on study cross-prediction performance (Fig. 2 ). This analysis showed that some studies generated high predictive power for CRC samples for their own study and in many instances performed well in external study prediction. In no case did training on any one study predict CRC uniformly well across all other studies. Failure of a study to predict samples in external studies may be due to a single outlier study. The factors contributing to study quality are variable and challenging to define. Despite the various shortcomings of meta-analyses, they are useful when optimizing algorithms that feature high generalizability. The focus of this report investigates feature selection methods and feature type (taxonomic and KO) used for modeling. We evaluated two distinct feature selection methods, FIRE and SIAMCAT. The best performing models generated by machine learning algorithms differ in terms of the features deemed to be of highest importance. Ensemble methods seek to make the best use of multiple models and their underlying selected features to establish both improved performance and increased generalizability [ 37 ]. We developed a workflow that included two distinct feature selection methods, FIRE and SIAMCAT, separately or in combination (Fig. 3 ). We used FIRE that aggregated features selected by the top 5 models. The optimal number of models used in ensemble approaches was not investigated in detail. We established the optimal number of taxonomic or gene features for each disease class. Surprisingly, the optimal number of features for CRC, both for taxonomic and gene features was substantially higher than that determined for CRAA and CRA ( Table 2 ). The reason(s) for this are not clear but may reflect that colorectal tumors have the largest effect on colonic microbiota and their encoded functions, thereby generating a larger spectrum of discriminatory biomarkers. For both CRAA and CRA, a relatively small number of KO features (40) was determined to be optimal. We speculate that this may reflect a relative paucity of discriminatory KO and taxonomic biomarkers at earlier stages of disease. We used SIAMCAT (Fig. 4 ) as an independent feature selection method that is based on statistical significance of features that we set at (p = < 0.001). While FIRE feature selection generally outperformed SIAMCAT, the value of SIAMCAT is evident from cases where the best performance was obtained by combining FIRE and SIAMCAT (Fig. 5 ). Furthermore, for all analyses involving non-redundant features derived from FIRE and SIAMCAT, SIAMCAT features were always present among the most important features positively contributing to external HO test AUC. Applying SIAMCAT to control and CRC samples generated a taxonomic feature importance list that is highly consistent with taxa reported by several studies [ 38 ], although the features identified through our process are more numerous than generally reported in other studies. CRC taxonomic features are unique as they are enriched for over-represented taxa and species normally resident in the oral cavity. The majority of studies examining these microbes have focused on their behavior in the oral cavity rather than the gut, but accumulating evidence suggests that these taxa are pathobionts capable of causing or contributing to disease in various contexts [ 39 , 40 ]. Perhaps most significant is the fact that among the top 24 CRC taxonomic features, all are oral bacteria. These organisms exist in the healthy gut microbiota, but are distinguished from CRC by their increase in relative abundance (fold-change) and an increase in prevalence. Examination of the top 50 CRC features, show that all but 3 taxonomic features display increased relative abundance in CRC relative to control healthy subjects. Among the esxceptions, two features, Roseburia intestinalis and members of the genus Anaerostipes are part of the normal commensal gut microbiota. The third case is unexpected involving Streptococcus salivarius , a known oral bacterium. The reason(s) for the decreased relative abundance of these taxa is unclear. The over-representation of oral microbes in CRC fecal samples is consistent with the idea that the tumor microenvironment co-selects these oral species through an unknown fitness advantage that is lacking in healthy individuals and/or a defense mechanism that becomes disabled in CRC. While the factors driving this fitness advantage may be complex, one factor that may explain these results is the metabolic shift occurring in colonic carcinoma epithelium over the transition from health and adenomas to carcinoma. This metabolic shift includes the replacement of oxygen consumption resulting from oxidative metabolism of butyrate for energy to non-oxygen consuming fermentation of lactate, referred to as aerobic glycolysis [ 41 ]. One important result of this metabolic shift is increased colonic oxygen tension at the site of the tumor. This increased oxygen content may be sufficient to positively select for the aerotolerant oral species observed. For all disease classes, the best performance on external test set AUC was obtained using a combination of taxonomic and KO features (Fig. 5 ). Upon examining the most important features for each disease class, we observed that external test set AUC for CRC was predominately driven by taxonomic features when a smaller number of features (< 50) is used, whereas the most important features for CRAA and CRA were biased for KO features (Fig. 6 ). We analyzed the overlap of features across disease classes and surprisingly found that when considering the direction of change, CRA and CRC features overlapping CRAA were very minimal and indeed primarily went in opposite directions, whereas a substantial number of features shared between CRA and CRC were altered in the same direction (Fig. 7 ). This result is most surprising and difficult to explain but suggests that the microenvironment and selective pressure of the gut is similar in CRA and CRC but diverges in CRAA. We hypothesize that the higher proportion of shared gene features relative to taxonomic features may reflect the functional redundancy of related and even distantly related taxa that by virtue of shared genes encoded in their respective genomes are essentially inter-changeable within the community. These results suggest two surprising conclusions concerning the selective microenvironments generated by colonic lesions and/or the factors produced by gut microbiota that drive disease onset and progression. First, the strong dissimilarity between CRA and CRAA microbiota suggests that advanced adenomas may not simply be larger forms of adenomas. Our results suggest instead indicate that the microbiota adenoma/advanced adenoma interactions represent distinct processes. Second, and perhaps even more surprising, is that given the strong opposing behavior of CRA and CRAA microbiota and gene representation, the CRA and CRC microbiota are more functionally synonymous. Among the 48 KO features that were differentially represented in all 3 disease classes compared to healthy controls, 41 (85%) displayed this pattern of agreement in direction between CRA and CRC and disagreement (15%) in direction between CRAA and the other disease classes. In this regard, the same gene features that are increased in common between CRA and CRC samples are decreased relative to healthy control samples in CRAA and vice-versa. This result is considered preliminary since nearly all of the advanced adenoma samples were derived from a single study. Additional samples will be required to determine whether these patterns of gene representation are generalizable. Conclusions and outlook. We have developed an ensemble method for feature selection that improves external test set performance of fecal microbiota AUC measures. We noted that meta-analysis of available datasets was associated with several potential confounders including study methods, read depth, and country of origin. Despite confounders and their limitations for meta-analysis, such data represent a powerful resource for developing high quality models that are generalizable. The implementation of FIRE and SIAMCAT displayed positive effects on diagnostic performance, although the best outcomes differed according to disease class, where FIRE alone sometimes outperformed features selected by FIRE and SIAMCAT. The top features generated by FIRE and SIAMCAT combined for CRC are highly consistent with multiple reports analyzing differential representation of taxa in CRC compared to healthy controls. Our results suggest that no single model or set of models are optimal for maximal external test set AUC for all three disease classes that instead require case by case optimization. Our results suggest that the development of an accurate, non-invasive diagnostic test using stool samples to identify all stages of CRC development is highly feasible and that the gut microbiome provides biomarkers that may outperform those used in other commercial CRC tests. Future studies should report pathology findings at higher resolution to address open questions such as whether sessile serrated polyps are associated with the same microbes as those harbored in subjects with hyperplastic or tubular adenomas. Additional studies featuring broadened geography and multiple studies from single countries will allow investigators to address whether biomarker variability is a country or study-specific effect. There is a relative paucity of studies examining fecal microbiota in subjects with CRA and CRAA. Additional studies are required to strengthen the diagnostic features associated with these disease states. Abbreviations AI artificial intelligence AUC area under the curve CPM counts per million CRA colorectal adenoma CRAA colorectal advanced adenoma CRC colorectal carcinoma DR data robot ETBF enterotoxin-producing Bacteroides fragilis FIRE feature importance rank ensembling FIT Fecal Immunochemical Test HO hold out KO KEGG ortholog ML machine learning NGS next generation sequencing PCoA Principal coordinate analysis SIAMCAT statistical inference of associations between microbial communities and phenotypes SNM supervised normalization method TMM trimmed mean of M-values Declarations Ethics approval and Consent to participate Not applicable Consent for publication All authors have given their consent for the publication. Availability of data and materials This study used data already available in public databases: Feng et al. 2015 https://www.ncbi.nlm.nih.gov/bioproject/PRJEB7774 Gao et al. 2021 https://www.ncbi.nlm.nih.gov/bioproject/PRJNA514108 https://www.ncbi.nlm.nih.gov/bioproject/PRJNA706060 Gupta et al 2019 https://www.ncbi.nlm.nih.gov/bioproject/PRJNA531273 Hanningan et al. 2018 https://www.ncbi.nlm.nih.gov/bioproject/PRJNA389927 Thomas et al. 2019 https://www.ncbi.nlm.nih.gov/bioproject/PRJNA447983 Vogtmann et al. 2016 https://www.ncbi.nlm.nih.gov/bioproject/PRJEB12449 Wirbel et al. 2016 https://www.ncbi.nlm.nih.gov/bioproject/PRJEB27928 Yachida et al. 2019 https://www.ncbi.nlm.nih.gov/bioproject/PRJDB4176 Spanoggianopoulis et al. 2022 https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA720145 Yu et al. 2017 https://www.ncbi.nlm.nih.gov/bioproject/PRJEB10878 Zeller et al. 2014 https://www.ncbi.nlm.nih.gov/bioproject/PRJEB6070 Competing interests S.N.P., A.E, P.K., G.K., E.F., Y.V., and T.K. are all employees of Prescient Metabiomics LLC, who are developing CRC stool diagnostic test. Other authors have no conflicts to report. Funding The study was supported by private investor funding to PMB Authors’ Contribution S.N.P. wrote the manuscript and conducted data analysis. A.M.E, P.Z.K., E.F., N.S., F.F., Y.V., and G.K. conceived and conducted data analysis. T.K. conceived and directed the study. Acknowledgements. The authors would like to thank Dan Knights and Emily Hollister for helpful discussion and critique of the manuscript and the staff at Diversigen for technical assistance. This work was supported by private funding provided to Prescient Metabiomics. References Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018;68:394–424. Lauby-Secretan B, Vilahur N, Bianchini F, Guha N, Straif K. The IARC Perspective on Colorectal Cancer Screening. N Engl J Med. 2018;378:1734–40. Lee JK, Liles EG, Bent S, Levin TR, Corley DA. Accuracy of Fecal Immunochemical Tests for Colorectal Cancer: Systematic Review and Meta-analysis. Ann Intern Med. 2014;160:171–81. Hundt S, Haug U, Brenner H. Comparative evaluation of immunochemical fecal occult blood tests for colorectal adenoma detection. Ann Intern Med. 2009;150:162–9. Imperiale TF, Ransohoff DF, Itzkowitz SH, Levin TR, Lavin P, Lidgard GP, et al. Multitarget Stool DNA Testing for Colorectal-Cancer Screening. N Engl J Med. 2014;370:1287–97. Frank C, Sundquist J, Yu H, Hemminki A, Hemminki K. Concordant and discordant familial cancer: Familial risks, proportions and population impact: Concordant and discordant familial cancer. Int J Cancer. 2017;140:1510–6. Francescangeli F, De Angelis ML, Zeuner A. Dietary Factors in the Control of Gut Homeostasis, Intestinal Stem Cells, and Colorectal Cancer. Nutrients. 2019;11:2936. Johnson CM, Wei C, Ensor JE, Smolenski DJ, Amos CI, Levin B, et al. Meta-analyses of colorectal cancer risk factors. Cancer Causes Control. 2013;24:1207–22. Huxley RR, Ansary-Moghaddam A, Clifton P, Czernichow S, Parr CL, Woodward M. The impact of dietary and lifestyle risk factors on risk of colorectal cancer: A quantitative overview of the epidemiological evidence. Int J Cancer. 2009;125:171–80. Garcia S, Pruitt SL, Singal AG, Murphy CC. Colorectal cancer incidence among Hispanics and non-Hispanic Whites in the United States. Cancer Causes Control. 2018;29:1039–46. Greathouse KL, White JR, Padgett RN, Perrotta BG, Jenkins GD, Chia N, et al. Gut microbiome meta-analysis reveals dysbiosis is independent of body mass index in predicting risk of obesity-associated CRC. BMJ Open Gastroenterol. 2019;6:e000247. Lee SH, Yun Y, Kim SJ, Lee E-J, Chang Y, Ryu S, et al. Association between Cigarette Smoking Status and Composition of Gut Microbiota: Population-Based Cross-Sectional Study. J Clin Med. 2018;7:E282. Rodriguez-Gonzalez A, Orio L. Microbiota and Alcohol Use Disorder: Are Psychobiotics a Novel Therapeutic Strategy? Curr Pharm Des. 2020;26:2426–37. Allen JM, Mailing LJ, Niemiro GM, Moore R, Cook MD, White BA, et al. Exercise Alters Gut Microbiota Composition and Function in Lean and Obese Humans. Med Sci Sports Exerc. 2018;50:747–57. Dai Z, Coker OO, Nakatsu G, Wu WKK, Zhao L, Chen Z, et al. Multi-cohort analysis of colorectal cancer metagenome identified altered bacteria across populations and universal bacterial markers. Microbiome. 2018;6:70. Thomas AM, Manghi P, Asnicar F, Pasolli E, Armanini F, Zolfo M, et al. Metagenomic analysis of colorectal cancer datasets identifies cross-cohort microbial diagnostic signatures and a link with choline degradation. Nat Med. 2019;25:667–78. Wirbel J, Pyl PT, Kartal E, Zych K, Kashani A, Milanese A, et al. Meta-analysis of fecal metagenomes reveals global microbial signatures that are specific for colorectal cancer. Nat Med. 2019;25:679. Arthur JC, Perez-Chanona E, Mühlbauer M, Tomkovich S, Uronis JM, Fan T-J, et al. Intestinal Inflammation Targets Cancer-Inducing Activity of the Microbiota. Science. 2012;338:120–3. Kostic AD, Chun E, Robertson L, Glickman JN, Gallini CA, Michaud M, et al. Fusobacterium nucleatum Potentiates Intestinal Tumorigenesis and Modulates the Tumor-Immune Microenvironment. Cell Host Microbe. 2013;14:207–15. Wu S, Rhee K-J, Albesiano E, Rabizadeh S, Wu X, Yen H-R, et al. A human colonic commensal promotes colon tumorigenesis via activation of T helper type 17 T cell responses. Nat Med. 2009;15:1016–22. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinforma Oxf Engl. 2010;26:139–40. Law CW, Chen Y, Shi W, Smyth GK. voom: Precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol. 2014;15:R29. Poore GD, Kopylova E, Zhu Q, Carpenter C, Fraraccio S, Wandro S, et al. Microbiome analyses of blood and tissues suggest cancer diagnostic approach. Nature. NIH Public Access; 2020;579:567. Mecham BH, Nelson PS, Storey JD. Supervised normalization of microarrays. Bioinforma Oxf Engl. 2010;26:1308–15. Wirbel J, Zych K, Essex M, Karcher N, Kartal E, Salazar G, et al. Microbiome meta-analysis and cross-disease comparison enabled by the SIAMCAT machine learning toolbox. Genome Biol. 2021;22:93. Yu J, Feng Q, Wong SH, Zhang D, Liang Q yi, Qin Y, et al. Metagenomic analysis of faecal microbiome as a tool towards targeted non-invasive biomarkers for colorectal cancer. Gut. BMJ Publishing Group; 2017;66:70–8. Feng Q, Liang S, Jia H, Stadlmayr A, Tang L, Lan Z, et al. Gut microbiome development along the colorectal adenoma–carcinoma sequence. Nat Commun. 2015;6:6528. Zeller G, Tap J, Voigt AY, Sunagawa S, Kultima JR, Costea PI, et al. Potential of fecal microbiota for early-stage detection of colorectal cancer. Mol Syst Biol. 2014;10:766. Yachida S, Mizutani S, Shiroma H, Shiba S, Nakajima T, Sakamoto T, et al. Metagenomic and metabolomic analyses reveal distinct stage-specific phenotypes of the gut microbiota in colorectal cancer. Nat Med. 2019;25:968–76. Gao R, Zhu Y, Kong C, Xia K, Li H, Zhu Y, et al. Alterations, Interactions, and Diagnostic Potential of Gut Bacteria and Viruses in Colorectal Cancer. Front Cell Infect Microbiol. 2021;11:657867. Gupta A, Dhakan DB, Maji A, Saxena R, P.K. VP, Mahajan S, et al. Association of Flavonifractor plautii, a Flavonoid-Degrading Bacterium, with the Gut Microbiome of Colorectal Cancer Patients in India. Bordenstein S, editor. mSystems. 2019;4:e00438-19, /msystems/4/6/msys.00438-19.atom. Vogtmann E, Hua X, Zeller G, Sunagawa S, Voigt AY, Hercog R, et al. Colorectal Cancer and the Human Gut Microbiome: Reproducibility with Whole-Genome Shotgun Sequencing. PLOS ONE. 2016;11:e0155362. Spanogiannopoulos P, Kyaw TS, Guthrie BGH, Bradley PH, Lee JV, Melamed J, et al. Host and gut bacteria share metabolic pathways for anti-cancer drug metabolism. Nat Microbiol. 2022;7:1605–20. Hannigan GD, Duhaime MB, Ruffin MT, Koumpouras CC, Schloss PD. Diagnostic Potential and Interactive Dynamics of the Colorectal Cancer Virome. mBio. 2018;9:e02248-18. Beghini F, McIver LJ, Blanco-Míguez A, Dubois L, Asnicar F, Maharjan S, et al. Integrating taxonomic, functional, and strain-level profiling of diverse microbial communities with bioBakery 3. eLife [Internet]. eLife Sciences Publications, Ltd; 2021 [cited 2022 Mar 31];10. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8096432/ Kanehisa M, Furumichi M, Tanabe M, Sato Y, Morishima K. KEGG: new perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res. 2017;45:D353–61. Bose S, Das C, Banerjee A, Ghosh K, Chattopadhyay M, Chattopadhyay S, et al. An ensemble machine learning model based on multiple filtering and supervised attribute clustering algorithm for classifying cancer samples. PeerJ Comput Sci. 2021;7:e671. Drewes JL, White JR, Dejea CM, Fathi P, Iyadorai T, Vadivelu J, et al. High-resolution bacterial 16S rRNA gene profile meta-analysis and biofilm status reveal common colorectal cancer consortia. NPJ Biofilms Microbiomes. 2017;3:34. Pushalkar S, Ji X, Li Y, Estilo C, Yegnanarayana R, Singh B, et al. Comparison of oral microbiota in tumor and non-tumor tissues of patients with oral squamous cell carcinoma. BMC Microbiol. 2012;12:144. Loftus M. Bacterial community structure alterations within the colorectal cancer gut microbiome. 2021;18. Wang G, Yu Y, Wang Y-Z, Wang J-J, Guan R, Sun Y, et al. Role of SCFAs in gut microbiome and glycolysis for colorectal cancer therapy. J Cell Physiol. 2019;234:17023–49. Additional Declarations Competing interest reported. S.N.P., A.E, P.K., G.K., E.F., Y.V., and T.K. are all employees of Prescient Metabiomics and developing a colo-rectal cancer diagnostic test. Supplementary Files FigureS1.pdf TableS1.xlsx Supplementary Table S1. Top 800 features list (by disease class). A full list of taxonomy and KO features derived from FIRE and SIAMCAT analysis). TableS2.xlsx Supplementary Tables S2. Taxonomic key for Figure 8. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-2838129","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Method Article","associatedPublications":[],"authors":[{"id":194041823,"identity":"c3fc71ad-cef7-475e-8105-17a7faba56af","order_by":0,"name":"Scott N. Peterson","email":"data:image/png;base64,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","orcid":"","institution":"Prescient Metabiomics","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Scott","middleName":"N.","lastName":"Peterson","suffix":""},{"id":194041824,"identity":"f815e864-476a-441a-90d2-e4a4359570c5","order_by":1,"name":"Alexey M. Eroshkin","email":"","orcid":"","institution":"Prescient Metabiomics","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Alexey","middleName":"M.","lastName":"Eroshkin","suffix":""},{"id":194041825,"identity":"17b47128-cb18-440e-8355-e3f388d3309c","order_by":2,"name":"Piotr Z. Kozbial","email":"","orcid":"","institution":"Prescient Metabiomics","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Piotr","middleName":"Z.","lastName":"Kozbial","suffix":""},{"id":194041826,"identity":"ebdd958b-94f0-451c-b465-4646ba65c714","order_by":3,"name":"Ermanno Florio","email":"","orcid":"","institution":"Prescient Metabiomics","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Ermanno","middleName":"","lastName":"Florio","suffix":""},{"id":194041827,"identity":"7afbe5fd-08aa-408f-8170-c220b09b5510","order_by":4,"name":"Farnaz Fouladi","email":"","orcid":"","institution":"Diversigen","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Farnaz","middleName":"","lastName":"Fouladi","suffix":""},{"id":194041828,"identity":"8651571f-9bc3-4fad-8148-7f9c080c4a59","order_by":5,"name":"Noah Strom","email":"","orcid":"","institution":"Diversigen","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Noah","middleName":"","lastName":"Strom","suffix":""},{"id":194041829,"identity":"62f52e58-36f4-4e90-8b9e-1c69bd27f22f","order_by":6,"name":"Yacgley Valdes","email":"","orcid":"","institution":"Prescient Metabiomics","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Yacgley","middleName":"","lastName":"Valdes","suffix":""},{"id":194041830,"identity":"cd07fb28-0f56-4c3e-9e11-c8aa0a4268fa","order_by":7,"name":"Gregory Kuehn","email":"","orcid":"","institution":"Prescient Metabiomics","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Gregory","middleName":"","lastName":"Kuehn","suffix":""},{"id":194041831,"identity":"dd3a30de-cd27-4023-95bb-3ca3f07262cb","order_by":8,"name":"Giorgio Casaburi","email":"","orcid":"","institution":"Exagen Inc","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Giorgio","middleName":"","lastName":"Casaburi","suffix":""},{"id":194041832,"identity":"b0bbb411-b59f-413b-b2b6-e15f853aae48","order_by":9,"name":"Thomas Kuehn","email":"","orcid":"","institution":"Prescient Metabiomics","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"","lastName":"Kuehn","suffix":""}],"badges":[],"createdAt":"2023-04-19 18:44:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2838129/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2838129/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":36279777,"identity":"fbd3a25b-3f81-4e0f-b2de-6dfbafa4fa6c","added_by":"auto","created_at":"2023-04-25 14:21:22","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":269151,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA. \u003c/strong\u003ePCoA plot of microbiota profiles derived from samples within the studies analyzed using TMM-Voom normalized data. \u003cstrong\u003eB. \u003c/strong\u003ePCoA plot of microbiota profiles derived from samples within the studies analyzed after supervised normalization (TMM-Voom-SNM).\u003cstrong\u003e C. \u003c/strong\u003ePCoA plot of microbiota profiles derived from samples for each disease class using TMM-Voom normalized data. \u003cstrong\u003eD. \u003c/strong\u003ePCoA plot of microbiota profiles derived from samples for each disease class after supervised normalization (TMM-Voom-SNM).\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/560c7865fe12f83af27728be.png"},{"id":36279226,"identity":"e219916a-2469-4ec3-8ae4-99ea1f3512a5","added_by":"auto","created_at":"2023-04-25 14:13:22","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":74197,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCross-correlation plot. \u003c/strong\u003eSamples from the studies listed at the top were used individually to train models for CRC. These models were used to predict HO samples (test set) from each study.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/6d2eadd05fce2e004d41e7b4.png"},{"id":36279228,"identity":"4dc731e9-6c90-49b9-a207-786999b7a4bf","added_by":"auto","created_at":"2023-04-25 14:13:22","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":73896,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eData Workflow.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/51ed874e1ab68fe496519fdb.png"},{"id":36278635,"identity":"15a57a83-3bf6-46c2-b83c-2b935dd45d1a","added_by":"auto","created_at":"2023-04-25 14:05:22","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":223152,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSIAMCAT. \u003c/strong\u003eFeatures are ranked according to significance scores. Box plots displaying the relative abundance of samples to visualize differential representation, fold-change, prevalence shift and feature contribution to AUC.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/96e9dcf2ee3c328e1419d65d.png"},{"id":36278632,"identity":"046dfd96-90f8-472b-9d8e-488d7b7a03a8","added_by":"auto","created_at":"2023-04-25 14:05:22","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":61058,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePerformance assessment. \u003c/strong\u003eTop achieved external test AUCs for models with different feature type and feature selection methods applied separately and together.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/fb458cea1eecc5cd3d7210fe.png"},{"id":36280483,"identity":"7ce9f025-3c1c-4e88-8d33-574a57f27c1a","added_by":"auto","created_at":"2023-04-25 14:29:22","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":98518,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVenn Diagrams. \u003c/strong\u003eThe number and % of total of overlapping features between disease classes is shown. The number of features corresponding to taxonomic (T) and gene features (K) are shown separately in red. The number of features analyzed are shown in decreasing order from left to right and top to bottom (800, 500, 200, 100, 50 and 20 features).\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/1e76f8303709974c02ccffae.png"},{"id":36281170,"identity":"4f68fdbb-be73-4d00-bfba-c7cd71921671","added_by":"auto","created_at":"2023-04-25 14:37:22","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":108619,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVenn Diagrams. \u003c/strong\u003eThe number of overlapping features for 800 taxonomic and gene features is shown in the center plot. The surrounding diagrams take into account the direction of change in a pairwise manner to illustrate similarities and differences between features common to disease classes. (FC)=fold-change.\u003c/p\u003e","description":"","filename":"Fig7.png","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/380e408ade2a4d6f40df1837.png"},{"id":36280493,"identity":"41ca5784-c0d6-452d-a194-638ab39cf791","added_by":"auto","created_at":"2023-04-25 14:29:23","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":283289,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDifferential representation of bacterial classes across disease classes. Figure 8. Differential representation of bacterial taxa across disease classes. \u003c/strong\u003eThe cladogram shows the phylogenetic placement of features that were informative. The outermost ring shows change in prevalence in CRA. Going inward, the next ring shows change in relative abundance for CRA followed by prevalence in CRAA, change in relative abundance for CRAA, prevalence in CRC, change in relative abundance for CRC. The volume of colored triangles is proportional to the magnitude of prevalence and fold change. Blue=negative change, red=positive change, green=negative change in prevalence, purple=positive change in prevalence. Taxonomic features follow the following labelling code described in the supplementary table 2.\u003c/p\u003e","description":"","filename":"Fig8.png","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/c2d8c236957b332398a9d312.png"},{"id":36713020,"identity":"0cbe590c-2c89-4f6a-adbd-56e6424f5990","added_by":"auto","created_at":"2023-05-08 17:29:30","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3072554,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/442ca3a4-5134-4001-bf14-0a29fab8fea3.pdf"},{"id":36278642,"identity":"077c36a5-a616-4582-a776-39667c605ca0","added_by":"auto","created_at":"2023-04-25 14:05:22","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":193950,"visible":true,"origin":"","legend":"","description":"","filename":"FigureS1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/2f2905cab89b03743a7a4a53.pdf"},{"id":36279778,"identity":"7e20a523-0eda-4718-b82c-872abfe867ac","added_by":"auto","created_at":"2023-04-25 14:21:22","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":155126,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table S1. Top 800 features list (by disease class). A full list of taxonomy and KO features derived from FIRE and SIAMCAT analysis).\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"TableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/9172fd64127413da5d902d63.xlsx"},{"id":36278640,"identity":"5c00a19e-17dc-42f0-80f3-b944451dbde3","added_by":"auto","created_at":"2023-04-25 14:05:22","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":12116,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Tables S2. Taxonomic key for Figure 8.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"TableS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-2838129/v1/9e94fb0ef9e85c9a379aa8ce.xlsx"}],"financialInterests":"Competing interest reported. S.N.P., A.E, P.K., G.K., E.F., Y.V., and T.K. are all employees of Prescient Metabiomics and developing a colo-rectal cancer diagnostic test.","formattedTitle":"Meta-analysis of Multi-functional Biomarkers for Discovery and Predictive Modeling of Colorectal Adenoma and Carcinoma","fulltext":[{"header":"Background","content":"\u003cp\u003eColorectal cancers are among the most prevalent cancers world-wide with an estimated 1.8\u0026nbsp;million new colon cancer cases and over 700,000 rectal cancer cases reported in 2018 [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Despite the strong evidence demonstrating that screening of individuals with average CRC risk reduces mortality [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], compliance amongst individuals is limited due to the invasiveness, discomfort and fear associated with colonoscopy. This has created a significant gap in CRC prevention within the health care system, emphasizing the need for sensitive, accurate, non-invasive diagnostic tests to detect colonic adenomas and carcinomas. It is expected that non-invasive tests will increase screening compliance over time. Our efforts seek to fill this gap by developing a powerful set of artificial intelligence (AI) driven methods to identify diagnostic biomarkers derived from microbial species present in stool samples.\u003c/p\u003e \u003cp\u003eAmong the non-invasive CRC detection tests is the Fecal Immunochemical Test (FIT) that is associated with limited sensitivity (79%) for detecting CRC [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] and a poor sensitivity (~\u0026thinsp;25%) for detecting advanced adenomas [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. A multi-target stool assay (Cologuard), measures KRAS mutations, aberrant NDRG4 and BMP3 methylation together with b-actin and hemoglobin immunoassays, performs better than FIT, detecting CRC cases with greater sensitivity (AUC\u0026thinsp;=\u0026thinsp;92% compared to 74% for FIT alone), however advanced premalignant lesions were still poorly detected by the Cologuard and FIT test assays (~\u0026thinsp;42% and ~\u0026thinsp;24% respectively) [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. These outcomes highlight another important gap in the healthcare system based on the relatively poor ability of existing non-invasive methods to detect early and advanced adenomas. Development of a diagnostic that addresses this gap will be of high impact both in terms of improved detection of pre-malignant lesions and potentially an overall reduction of colonoscopies required for average risk subjects.\u003c/p\u003e \u003cp\u003eCRC is a heterogeneous disease, the majority of cases are sporadic without underlying heritable features [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. A wide variety of environmental factors including a western diet, obesity, cigarette smoking, alcohol consumption and lack of exercise are known CRC risk factors [\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Chief amongst these risk factors is diet, that was linked to an estimated\u0026thinsp;~\u0026thinsp;38% of incipient CRC cases. Additional evidence for environmental influence of CRC is based on findings that the incidence of CRC is influenced by emigration, wherein a subject\u0026rsquo;s risk of CRC development is altered based on the diet and lifestyle of the recipient country [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Each of the above-mentioned CRC risk modifiers is also known to modulate the composition of the gut microbiota [\u003cspan additionalcitationids=\"CR12 CR13\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. This association has drawn substantial attention to the gut microbiota as a potential mediator of CRC initiation and/or progression, and accumulating evidence supports this hypothesis. The large number of species and genes encoded in the gut microbiome represents an attractive source of potential biomarkers for diagnostics and prognostics of early, premalignant adenomas, advanced adenomas, and CRC.\u003c/p\u003e \u003cp\u003eSeveral studies have examined gut microbiota in CRC using either 16S rRNA or shotgun metagenomic sequencing. These studies have explored fecal and mucosal-associated microbial populations and different stages along the adenoma, carcinoma progression. One previous meta-analysis of fecal microbiota datasets resulted in the identification of seven bacterial species enriched in CRC (\u003cem\u003eBacteroides fragilis, Fusobacterium nucleatum, Parvimonas micra, Porphyromonas assacharolytica, Prevotella intermedia, Alistipes finegoldii\u003c/em\u003e and \u003cem\u003eThermoanaeroovibrio acidaminovorans\u003c/em\u003e [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. A separate pair of meta-analyses identified an expanded set of twenty-nine species enriched over eight distinct geographical regions [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. A number of studies have analyzed the human gut microbiota associated with colonic tumors and normal adjacent tissue, leading to the identification of dysbiotic signatures associated with CRC. While specific taxa vary from study to study, some common themes have emerged including the frequent identification of elevated relative abundance of \u003cem\u003eE. coli\u003c/em\u003e [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], \u003cem\u003eFusobacterium nucleatum\u003c/em\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] and enterotoxin-producing \u003cem\u003eBacteroides fragilis\u003c/em\u003e (ETBF) strain [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Additional taxa associated with CRC have also been identified, but are less uniformly observed across studies.\u003c/p\u003e \u003cp\u003eAn important observation established by these studies is that the magnitude of differentiation in relative abundance of microbial biomarkers derived from tissue samples is substantially greater than that of stool samples. In stool samples, differentially abundant taxa are more subtle, often requiring AI-based methods for detection. A number of characteristics of fecal microbiota present specific challenges in the identification of diagnostic biomarkers for early detection of adenomas and carcinomas, including high dimensionality, data sparsity, and low generalizability. Despite the massive quantity of DNA sequence data generated in shotgun metagenomic sequencing of stool samples, we have observed that the best performing taxonomic biomarkers are detected in only a relatively small number of samples. This exemplifies the problem of data sparsity and dictates that a high-performance diagnostic test based on next generation sequencing (NGS) sequence data will require multiple independent biomarkers to compensate for low prevalence of any single microbial biomarker in the human population.\u003c/p\u003e \u003cp\u003eHere we describe a series of methodological improvements to a diagnostic NGS data analysis pipeline that addresses these challenges. Our pipeline automates several steps that increase efficiency and performance of available models, ultimately improving the power of taxonomic and KEGG ortholog (KO), gene features to distinguish healthy subjects from those with early and advanced adenomas and those with carcinomas. An overarching goal of this study was to improve the generalizability of machine-learning models that avoid a variety of known pitfalls associated with metagenomic data, These pitfalls include the use of heterogeneous methods and analysis procedures for data generation. Our analysis of metagenomic datasets indicate distinctions in the most informative biomarkers for each disease class. Features of highest importance distinguishing CRC from healthy controls were disproportionately reliant on taxonomic features, whereas CRA and CRAA features were more balanced in representation of KO and taxonomic features. We note that the optimal features for each disease class display very little overlap, suggesting that the adenoma-to-carcinoma progression involves uniquely selective biochemical environments for fecal microbiota that do not follow a simple linear relationship. These results are discussed with an outlook toward future studies, that may facilitate increased diagnostic power, deciphered from fecal microbiota.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eData processing.\u003c/strong\u003e The raw fecal shotgun DNA sequence data was preprocessed and taxonomically profiled using the bioBakery 3 pipeline v3.0.0-alpha.7 using the default parameters and kneaddata v0.10.0, MetaPhlAn version 3.0.7, HUMAnN v3.0.0.alpha.4, biobakery_workflows. Functional profiling of metagenomic sequencing data was performed using HUMAnN 3 genefamilies.tsv output and KEGG orthology gene groups (KO). To facilitate data interpretation and visualization, we regrouped gene family abundance data (represented by UniRef identifiers) into KO functional categories using humann_regroup_table script.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData normalization and visualization by PCoA.\u003c/strong\u003e Total sequence read counts including taxa from all ranks (kingdom to species) were normalized using weighted trimmed mean of M-values (TMM) using the \u0026lsquo;edgeR\u0026rsquo; package[\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e] and converted into log-counts per million (log-CPM) using the \u003cem\u003evoom\u003c/em\u003e function implemented in the \u0026lsquo;limma\u0026rsquo; package [\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e] in R version 4.2.1. The data were further normalized using a supervised normalization method (SNM) as described [\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e] to remove significant batch effects between projects while retaining biological differences between disease classes. The SNM method was implemented in the \u0026lsquo;snm\u0026rsquo; package [\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e] in R. The effects of supervised normalization were visualized using Principal Coordinate Analysis (PCoA). PCoA was performed using Euclidean distances on both TMM-Voom and TMM-Voom-SNM transformed count tables. Differences in the microbial composition between projects and disease class were assessed separately using the \u003cem\u003eadonis2\u003c/em\u003e function from the \u0026lsquo;vegan\u0026rsquo; package (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://CRAN.R-project.org/package=vegan\u003c/span\u003e\u003c/span\u003e). Mean distance of samples from the centroids in the PCoA plots was compared between projects using the Kruskal-Wallis test.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMachine learning.\u003c/strong\u003e Models were created using the automated machine learning platform called DataRobot (DR); (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.datarobot.com\u003c/span\u003e\u003c/span\u003e). A custom python script was developed for automated data submission to DR that allows the development of models for multiple datasets. The best model of all developed models was selected based on the largest area under the curve (AUC) value for external test dataset prediction. \u0026ldquo;Blender models\u0026rdquo; which are obtained using several machine learning algorithms (combining the predictions of two or more models), were not used here. For classification purpose for each disease target, the set of samples was divided randomly into a training set (80% of samples) and a test set (20% of samples). The training set was used to develop a set of high performing predictive models in DR (using more than ten different machine learning architectures for classification such as eXtreme Gradient Boosted Trees Classifier, Keras Slim Residual Neural Network Classifier using Training Schedule, Elastic-Net Classifier and Light Gradient Boosted Trees Classifier with Early Stopping). The developed models were used to predict the disease state in the remaining 20% of samples (unseen data which were not used in training), and the top model (with highest external test AUC) is defined. The model performance on the test set parameters was then supplemented with external test sensitivity, specificity, and accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDataRobot feature lists.\u003c/strong\u003e Feature lists control the subset of features that DataRobot used to build models. DataRobot automatically creates several feature lists for each project, including Informative Features and DR Reduced Features. Informative Features are all features that provide information potentially valuable for modeling (normally all features). DR Reduced Features are a subset of features, selected based on the Feature Impact calculation of the best model. The DR Reduced feature list consisted of the features that provide 95% of the accumulated impact for the model. Since Informative Feature lists usually have almost all features in the dataset (1,000\u0026ndash;2,000 in taxonomy annotation and 5,000\u0026ndash;10,000 in functional annotation) and DR Reduced feature list have no more than 100 features, for comparative purposes we used models built using DR Reduced feature lists.\u003c/p\u003e\n\u003cp\u003ePermutation-based Feature Impact measures a drop in model accuracy when a feature\u0026rsquo;s values are shuffled. To compute these values, DataRobot makes predictions on a sample of training records and then alters the training data (shuffles value) computing a drop in accuracy that resulted from shuffling. This shuffling process was repeated and the results were normalized and ranked (top feature has an impact of 100%). The sampling process corresponds to one of the following criteria: For balanced data, random sampling is used. For imbalanced binary data, smart down sampling is used; DataRobot attempts to make the distribution for imbalanced binary targets closer to 1:1 and adjusts the sample weights used for scoring. For zero-inflated regression data, smart down sampling was used; DataRobot groups the non-zero elements into the minority class. For imbalanced multiclass data, random sampling was used.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFIRE feature selection.\u003c/strong\u003e A feature reduction and selection method \u0026ldquo;Feature Importance Rank Ensembling\u0026rdquo; (FIRE, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.datarobot.com/blog/using-feature-importance-rank-ensembling-fire-for-advanced-feature-selection/\u003c/span\u003e\u003c/span\u003e) was used. In this method, the features are derived from multiple diverse predictive models built by DR. These models were sorted by external test AUC. The median rank of each feature was calculated by aggregating the ranks for each of the top models (the number of top models to consider was empirically selected equal to five).\u003c/p\u003e\n\u003cp\u003eThe FIRE procedure consists of the following steps: (a) calculate the feature importance for the top 5 models (determined by the external test AUC), (b) obtain the ranking of the features, (c) compute and sort the aggregated list by the computed median rank, (d) define the threshold number of features to select, (e) define a feature list based on the newly selected features.\u003c/p\u003e\n\u003cp\u003eSince the optimal number of features is not known, we iteratively tested several thresholds with large increments in the first loop (800, 700, 600, 500, 400, 300, 200, 100, 90, 80, 70, 60, 50, 40, 30) and small increments around the first found threshold in the second loop (for example, 85, 84, 83, 82, 81, 79, 78, 77, 76, 75, if the best threshold from the first loop equals 80). We took the threshold that provided the highest external test AUC. We considered the maximal number of FIRE features as 800 since in some of the analyzed datasets, the total feature number was between 800 and 900. We implemented FIRE in a multi-step process to exploit ensemble approaches to establish feature selection with increased generalizability since its basis is not tied to a single model and its associated biases.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSIAMCAT feature selection.\u003c/strong\u003e Another feature selection method used here is based on statistical inference of associations between microbial communities and phenotypes and is referred to as SIAMCAT version 2.1.0 [\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e]. SIAMCAT is part of the suite of microbiome analysis tools developed at EMBL. We applied the cut-off of adjusted p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.001) to identify important features.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCombining FIRE and SIAMCAT.\u003c/strong\u003e Since the FIRE and SIAMCAT feature lists are selected based on different criteria (ensemble ranking for the first and statistical significance for the second), we also tested feature lists that combine the FIRE and SIAMCAT features together to create new feature list.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eTo evaluate the accuracy and feasibility of developing a non-invasive diagnostic stool test for early and advanced pre-cancerous adenomas and carcinomas, we imported data generated from 11 studies conducted by laboratories in 9 countries, analyzing stool samples by shotgun metagenomic sequencing [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e34\u003c/span\u003e]. All samples were confirmed by colonoscopy. The descriptive statistics of the cohorts from each study are shown in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u0026nbsp;\u003c/p\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003e\u003cstrong\u003eSelected Metagenomic projects used for modeling.\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eProject\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNumber\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGender\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCountry\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003edisease type\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eraw reads\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003epost-qc reads\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e(%)\u003c/p\u003e\n \u003cp\u003ehuman reads\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSpanogiannopoulis (34)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMale:5 female 5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e52\u0026thinsp;\u0026plusmn;\u0026thinsp;20.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMissing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eUSA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 0 CRC: 10 CRA: 0 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e37213448\u0026thinsp;\u0026plusmn;\u0026thinsp;6907944\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e34520932\u0026thinsp;\u0026plusmn;\u0026thinsp;6316395\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.00\u0026thinsp;\u0026plusmn;\u0026thinsp;9.49\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFeng (27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e156\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 88 female: 68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e66.9\u0026thinsp;\u0026plusmn;\u0026thinsp;8.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e27.4\u0026thinsp;\u0026plusmn;\u0026thinsp;4.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAustria\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 63 CRC: 46 CRA: 0 CRAA: 47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e52689474\u0026thinsp;\u0026plusmn;\u0026thinsp;8343659\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e46088635\u0026thinsp;\u0026plusmn;\u0026thinsp;7292627\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.63\u0026thinsp;\u0026plusmn;\u0026thinsp;1.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGao (30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e126\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMissing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMissing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMissing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChina\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 47 CRC: 39 CRA: 40 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e46462323\u0026thinsp;\u0026plusmn;\u0026thinsp;16612805\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e42959333\u0026thinsp;\u0026plusmn;\u0026thinsp;15584240\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.45\u0026thinsp;\u0026plusmn;\u0026thinsp;3.55\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGupta (31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 18 female: 11 missing: \\ 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e59.8\u0026thinsp;\u0026plusmn;\u0026thinsp;7.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMissing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIndia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 0 CRC: 30 CRA: 0 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e9229167\u0026thinsp;\u0026plusmn;\u0026thinsp;4142109\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e8510190\u0026thinsp;\u0026plusmn;\u0026thinsp;3816170\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.7\u0026thinsp;\u0026plusmn;\u0026thinsp;0.651\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHannigan (35)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 46 female: 35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e58.6\u0026thinsp;\u0026plusmn;\u0026thinsp;10.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28.1\u0026thinsp;\u0026plusmn;\u0026thinsp;6.1 missing: 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCanada, USA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 28 CRC: 27 CRA: 26 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6593685\u0026thinsp;\u0026plusmn;\u0026thinsp;3784609\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4964982\u0026thinsp;\u0026plusmn;\u0026thinsp;2801436\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.69\u0026thinsp;\u0026plusmn;\u0026thinsp;0.718\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eThomas (16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e140\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 52 female: 28 missing: 60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e67.5\u0026thinsp;\u0026plusmn;\u0026thinsp;8.73 missing: 60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25.5\u0026thinsp;\u0026plusmn;\u0026thinsp;3.93 missing: 64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eItaly\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 52 CRC: 61 CRA: 27 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e44984798\u0026thinsp;\u0026plusmn;\u0026thinsp;24403021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e41938748\u0026thinsp;\u0026plusmn;\u0026thinsp;22955372\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.52\u0026thinsp;\u0026plusmn;\u0026thinsp;0.501\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVogtmann (33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e104\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 74 female: 30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e61.5\u0026thinsp;\u0026plusmn;\u0026thinsp;12.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25.1\u0026thinsp;\u0026plusmn;\u0026thinsp;4.25 missing: 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eUSA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 52 CRC: 52 CRA: 0 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e62406634\u0026thinsp;\u0026plusmn;\u0026thinsp;15463669\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e55272649\u0026thinsp;\u0026plusmn;\u0026thinsp;14006526\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.62\u0026thinsp;\u0026plusmn;\u0026thinsp;2.51\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWirbel (17)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e130\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 76 female: 54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e63.4\u0026thinsp;\u0026plusmn;\u0026thinsp;12.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e24.9\u0026thinsp;\u0026plusmn;\u0026thinsp;4.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGermany\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 60 CRC: 70 CRA: 0 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e25277871\u0026thinsp;\u0026plusmn;\u0026thinsp;9126431\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e23317638\u0026thinsp;\u0026plusmn;\u0026thinsp;8560313\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.75\u0026thinsp;\u0026plusmn;\u0026thinsp;0.791\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYachida (29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e611\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 353 female: 258\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e61.8\u0026thinsp;\u0026plusmn;\u0026thinsp;11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e22.9\u0026thinsp;\u0026plusmn;\u0026thinsp;3.37 missing: 10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eJapan\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 286 CRC: 258 CRA: 67 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e45765841\u0026thinsp;\u0026plusmn;\u0026thinsp;12910710\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e41600333\u0026thinsp;\u0026plusmn;\u0026thinsp;11694276\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.13\u0026thinsp;\u0026plusmn;\u0026thinsp;0.576\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYu (26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 81 female: 47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e64.2\u0026thinsp;\u0026plusmn;\u0026thinsp;9.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23.8\u0026thinsp;\u0026plusmn;\u0026thinsp;3.08 missing: 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eChina\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 54 CRC: 74 CRA: 0 CRAA: 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e56317665\u0026thinsp;\u0026plusmn;\u0026thinsp;9956025\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e48374964\u0026thinsp;\u0026plusmn;\u0026thinsp;9470724\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.35\u0026thinsp;\u0026plusmn;\u0026thinsp;1.63\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eZeller (28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e154\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale: 84 female: 70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e63.1\u0026thinsp;\u0026plusmn;\u0026thinsp;12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25.5\u0026thinsp;\u0026plusmn;\u0026thinsp;4.04 missing: 4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFrance, Germany\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCTR: 61 CRC: 91 CRA: 1 CRAA: 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e58257017\u0026thinsp;\u0026plusmn;\u0026thinsp;23112145\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e50340355\u0026thinsp;\u0026plusmn;\u0026thinsp;20826272\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.21\u0026thinsp;\u0026plusmn;\u0026thinsp;5.54\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eDescriptive statistics of gender, BMI, age, disease classification, raw reads, post-qc reads, and percentage of human reads for each project. For continuous variables, mean and standard deviation are shown and for categorical variables number of samples within each category is shown.\u003c/p\u003e\n\u003cp\u003eTo analyze publicly available data sets we first performed data normalization as described [\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e] and (see methods). The effects of supervised normalization were visualized using Principal Coordinate Analysis (PCoA). Differences in the microbial composition between projects and disease class were assessed separately. PCoA plots showed that supervised normalization significantly reduced the variation that could be explained by unique projects from an R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;10\u0026ndash;0.075% (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eA-B). Only 1.4% of the variation was explained by disease class using the TMM-Voom normalized data (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eC). This variation decreased to 0.896% following supervised normalization, though the difference between disease types remained significant (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eD).\u003c/p\u003e\n\u003cp\u003eTo identify projects and samples representing potential outliers we determined distance to centroids within each project. These distances were similar across projects for both TMM-Voom and TMM-Voom-SNM data (\u003cstrong\u003eSupplementary Figure \u003cspan class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eAlthough non-ideal, the project-specific variation was fully expected. We elected to conduct performance optimization of these datasets rather than attempting to remove studies based on \u003cem\u003ead hoc\u003c/em\u003e criteria. These results highlight the significant challenges associated with meta-analyses of gut microbiome data. It is difficult to distinguish between study and country-specific effects. Therefore, the possibility that biomarkers associated with adenomas and carcinomas are prone to country or regional-specific effects remains unresolved.\u003c/p\u003e\n\u003cp\u003eTo illustrate study-variability we used data from individual studies to train models which were then used to predict samples from all other studies to determine how well each model predicted external samples (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eIn most instances, study-specific models performed well on themselves. Some studies used for training generated relatively higher AUC for test sets across all or most studies. Other studies used as training sets predicted one or a few studies with high AUC but displayed greater variation overall. Finally, some studies performed relatively poorly as training sets for most or all other studies. The reasons for cross-study variability may be numerous and include sampling differences, methodological variability, read-depth, geographic effects, lesion location, cohort demographics, and others. This result may be a way to measure the generalizability of features derived from particular studies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature generation.\u003c/strong\u003e In our efforts to develop a stool microbiome diagnostic analysis pipeline, we have focused on the evaluation of two related data features. The first, are the relative abundance/prevalence of taxonomic features [\u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e]. Second, we explored the inclusion of gene features derived from shotgun metagenomic sequence analysis. The KEGG Ortholog (KO) grouping is a database of molecular functions representing functional orthologs [\u003cspan class=\"CitationRef\"\u003e36\u003c/span\u003e]. We tested the hypothesis that KO features may positively contribute to predictive performance when combined with those derived from taxonomy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature Processing and Selection.\u003c/strong\u003e We have implemented strategies to evaluate a large variety of feature reduction methods to compare their overall impact on prediction accuracy. Each of these had specific strengths and weaknesses. We expected that feature selection schemes based on filtering low-prevalence features would be risky in the context of fecal microbiota since many of the best diagnostic features (species) are of low abundance and/or low prevalence. Low prevalence dictates that the best performing models are likely to require a larger number of features for optimal accuracy. Here we evaluate the feature reduction method referred to as Feature Importance Rank Ensembling (FIRE) (see methods), in which the highest-ranking features are derived from multiple diverse models. We elected to identify the best 5 models (by an external test AUC) in our ensemble procedures. Five models were chosen empirically since a general degradation in performance was observed in the sixth best model. We added another feature selection method based on statistics referred to as SIAMCAT to define a novel workflow (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eOne important finding based on implementation of FIRE is that the best performing models differ according to disease class, emphasizing that no single model or set of models is optimal to distinguish health from CRA, CRAA, CRC. We therefore performed FIRE using feature selection and training on CRA, CRAA and CRC samples independently to achieve target-specific model optimization. This in turn achieves optimal disease-class-specific diagnostic performance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFIRE Feature Selection.\u003c/strong\u003e While our results were based on a process that began with 800\u0026ndash;1100 features, in practice, any number of features may be selected, however an optimal feature number must be determined empirically. To establish the optimal number of features for each disease class, FIRE was performed iteratively to establish a performance score based on external test AUC. We conducted these analyses for each disease target and for taxonomic and KO features separately (\u003cstrong\u003eTable\u0026nbsp;2\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable\u0026nbsp;2. FIRE Feature Selection.\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"1263\" height=\"506\"\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eThe table reports AUC values for the external (20%) data sets. Bold underlined values are the maximal external test AUC achieved for a particular annotation, target, and FIRE features set size. \u003csup\u003e1\u003c/sup\u003e eXtreme Gradient Boosted Trees Classifier, \u003csup\u003e2\u003c/sup\u003e Keras Slim Residual Neural Network Classifier using Training Schedule (1 Layer: 64 Units), \u003csup\u003e3\u003c/sup\u003e Elastic-Net Classifier (L2 / Binomial Deviance), \u003csup\u003e4\u003c/sup\u003e Light Gradient Boosted Trees Classifier with Early Stopping\u003c/p\u003e\n\u003cp\u003eWe did not observe any pattern across disease classes when evaluating taxonomic features and functional KO features separately. For CRC, 800 KO features and 400 taxonomic features provided the best performance. This was strongly contrasted by CRAA and CRA analyses. For CRAA the optimum number of KO and taxonomic features was substantially lower (40 and 100, respectively). For CRA we observed 40 KO features and 70 taxonomic features were optimal.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSIAMCAT.\u003c/strong\u003e As an additional layer of ensemble-based analysis, we processed taxonomic and gene features through SIAMCAT to allow visualization of differential abundance, prevalence, feature AUC and feature ranking based on statistical significance (Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eThe feature importance of taxa associated with CRC (adj p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), illustrate that some taxonomic features display negligible fold-change, but display significant shifts in prevalence. This characteristic was most pronounced in CRC microbiota.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCombining FIRE and SIAMCAT features.\u003c/strong\u003e We combined features generated by FIRE and SIAMCAT. Not surprisingly, the features generated by FIRE and SIAMCAT partially overlap. The unique features from a combined list were used to classify samples into healthy or disease classes. To determine whether combining FIRE and SIAMCAT selected features improves performance we evaluated possible incremental improvements of our approach by generating AUCs using the best model from machine learning algorithms to establish a baseline for comparison to FIRE and SIAMCAT alone and in combination (Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eThe best performance for CRA was achieved when combining taxonomic and KO features selected by combined FIRE-SIAMCAT. This approach yielded a nearly 8% increase in external AUC (baseline AUC\u0026thinsp;=\u0026thinsp;0.80 vs 0.87). Analysis of CRAA performance was somewhat more complex. All feature types performed best when using FIRE alone and there was little difference in the performance when using taxonomic features alone or in combination with KO features. FIRE selected features generated a 3% gain in external AUC (baseline AUC\u0026thinsp;=\u0026thinsp;0.94 vs 0.97). The results for CRC showed that the combination of taxonomic and KO features outperformed taxonomic features alone which in turn outperformed KO features alone. The best performance was observed from features selected by FIRE, resulting in a modest 2% increase in external AUC (baseline AUC\u0026thinsp;=\u0026thinsp;0.80 vs 0.82). A major conclusion from this analysis is that the microbiota and microbiome associated with each disease class demand defining distinct computational workflows, as no single or set of models perform optimally on all 3 disease classes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature overlap across disease classes.\u003c/strong\u003e To gain biological insights into the features that contribute most significantly to distinguishing or unifying disease class prediction, we analyzed 800 features generated from a combination of FIRE and SIAMCAT across disease classes (Fig. \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e). Figure \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eWhen comparing the top 20 features (bottom right Venn diagram), there was no overlap in either taxonomic or gene features. It is notable that 60% of the features for CRC are taxonomic, significantly larger than that observed for CRA (40%) and CRAA (25%). When examining the top 50 features (bottom middle Venn diagram), differences in the proportion of features derived from taxa or KO dissipate. Among the top 50 features we observe modest overlap in features across disease classes. Comparison of the top 100 features shows that the proportions of taxonomic features become quite even across disease classes. As more features are compared, we note that the proportion of gene features continue to increase relative to taxonomic features, and we observe increasing overlap across disease classes. Examination of 800 features reveals that among the overlaps, KO features dominate relative to taxonomic features. Indeed, we observed 48 KO features (2.4%) shared among all 3 disease classes and no cases of overlapping taxonomic features. This imbalance is also evident in all pairwise comparisons of overlapping features such that taxonomic features represent between ~\u0026thinsp;9\u0026ndash;14% of overlapping features. Given that shared taxonomic features frequency is similar across disease classes, it is notable that the number of shared gene features is significantly higher between CRC and CRA relative to any other pairwise relationship.\u003c/p\u003e\n\u003cp\u003eTo refine these comparisons, we considered the direction of change of feature types (increased or decreased in disease), to better approximate the biological similarity of shared taxonomic and gene features (Fig. \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003eThe diagram (top left) indicates that the CRA and CRC microbiome share a significantly higher number of features altered in relative abundance in the same direction compared to other pair-wise comparisons. The relationship between CRA and CRAA samples (top right) shows the small number of features displaying the same direction of change. We note that the large number of overlapping features shared between CRC vs CRAA and CRA vs CRAA) display the opposite direction of change. This result suggests that a subset of features contribute to prediction performance in all disease targets however, the direction of change of these common features in fact, distinguishes CRAA from CRA and CRC.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel validation: analysis of taxonomic features.\u003c/strong\u003e A comparison of the features established using our ensemble feature selection approach (\u003cstrong\u003eSupplementary Table \u003cspan class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/strong\u003e) to other similar large-scale meta-analyses confirmed a subset of features associated with CRC including; \u003cem\u003eFusobacterium nucleatum\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e], \u003cem\u003eGemella morbillorum\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e], \u003cem\u003eBifidobacterium catenulatum\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e], \u003cem\u003ePeptostreptococcus anaerobius\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e], \u003cem\u003ePeptostreptococcus stomatis\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e], \u003cem\u003ePorphyromonas asaccharolytica\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e], \u003cem\u003eParvimonas micra\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e], \u003cem\u003eSolobacterium\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e], \u003cem\u003eClostridium symbiosum\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e], \u003cem\u003eEubacterium eligens\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e], \u003cem\u003ePrevotella intermedia\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e], \u003cem\u003eHungatella hathewayi\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e], \u003cem\u003eRoseburia intestinalis\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e], \u003cem\u003eFusobacterium sp oral taxon 370\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e], \u003cem\u003eBacteroides fragilis\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e], \u003cem\u003ePorphyromonas uenonis\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e], \u003cem\u003ePrevotella nigrescens\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e] and \u003cem\u003eAnaerococcus vaginalis\u003c/em\u003e [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e]. These features may represent the most \u0026ldquo;universal\u0026rdquo; biomarkers for CRC that over-ride confounding country-specific effects, although other aspects of study design and technical methods and reporting are also likely to explain why more features are not conserved across studies.\u003c/p\u003e\n\u003cp\u003eThe best predictive KO features (\u003cstrong\u003eSupplementary Table \u003cspan class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/strong\u003e) did not provide biological insight, since under- and over-represented KO\u0026rsquo;s taken from the top 800 features (KO and taxonomy) did not generate sufficient coverage of pathways to establish confidence in their potential impact on human physiology. Among the 735 KO features associated with CRA and 713 associated with CRAA, only 10 (1.4%) were altered in the same direction, whereas 122 (16.6%) were altered in opposite direction. Similarly, a comparison of the 704 KO features associated with CRC, 28 (4%) displayed the same direction of change as that observed in CRAA, whereas 108 (15.3%) were in opposite direction. Finally, a comparison of KO features in common between CRA and CRC showed that 25.4% changed in the same direction whereas only 1.7% changed in the opposite direction. We are currently conducting a comprehensive global analysis of gut microbiome functional features (in preparation) and therefore describe only taxonomic features here.\u003c/p\u003e\n\u003cp\u003eIn order to assess whether taxonomic features for each class represent coherent phylogenetic groups and direction of change, we analyzed important features at various phylogenetic levels. Among the top 800 features, 65 represented taxa over- or under-represented in CRA, 87 taxa for CRAA and 96 taxa for CRC. It should be noted that these features generally did not achieve statistical significance in comparisons but were deemed discriminatory based on AI models used. In total, the feature importance list contained taxa from 14 classes. The top features for all disease classes defined 41 different bacterial families.\u003c/p\u003e\n\u003cp\u003eThe distribution of important features for each disease class were disproportionately associated with particular phylogenetic groups (Fig. \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e\u003cstrong\u003eand Supplementary Table \u003cspan class=\"InternalRef\"\u003eS2\u003c/span\u003e\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eFor CRAA, within the phylum, Archaea, \u003cem\u003eMethanobrevibacter smithii\u003c/em\u003e is uniquely over-represented compared to healthy controls. Within the phylum, Actinobacteria, taxa over-represented in CRAA samples were disproportionately represented. Within the family Actinomycetaceae, \u003cem\u003eA. graevenitzii, Actinomyces\u003c/em\u003e sp HMSC035G02, \u003cem\u003eActinomyces\u003c/em\u003e sp ICM47, \u003cem\u003eA. viscosus\u003c/em\u003e were all over-represented in CRAA samples. In CRA samples \u003cem\u003eA. odontolyticus\u003c/em\u003e was under-represented, whereas \u003cem\u003eActinomyces cardiffensis\u003c/em\u003e and \u003cem\u003eActinomyces turicensis\u003c/em\u003e were over-represented in CRC samples. Among the Bifidobacteriaceae, only 2 species ranked as informative, \u003cem\u003eB. longum\u003c/em\u003e is over-represented in CRAA samples, whereas \u003cem\u003eB. catenulatum\u003c/em\u003e, is under-represented in CRC samples. Among the Propionibacteriaceae, \u003cem\u003eP. freudenreichii\u003c/em\u003e was uniquely over-represented in CRAA samples. Within the family Atopobiaceae, \u003cem\u003eAtopobium rimae\u003c/em\u003e and \u003cem\u003eOlsenella scatoligenes\u003c/em\u003e were both over-represented in CRAA samples. Within the family Coriobacteriaceae, \u003cem\u003eCollinsella intestinalis\u003c/em\u003e, \u003cem\u003eCollinsella stercoris\u003c/em\u003e and \u003cem\u003eEnorma massiliensis\u003c/em\u003e and [\u003cem\u003eCollinsella\u003c/em\u003e] \u003cem\u003emassiliensis\u003c/em\u003e were all over-represented in CRAA samples, whereas the latter was under-represented in CRC samples. Several genera within Eggerthellaceae were over-represented in CRAA samples including; \u003cem\u003eAdlercreutzia, Asaccharobacter, Enterorhabdus\u003c/em\u003e and \u003cem\u003eGordonibacter\u003c/em\u003e were over-represented in CRAA samples, whereas the latter two were under-represented in CRA samples. \u003cem\u003eSlackia exigua\u003c/em\u003e was uniquely over-represented in CRC samples.\u003c/p\u003e\n\u003cp\u003eThe 11 differentially represented \u003cem\u003eBacteroides\u003c/em\u003e spp., were distributed across all disease classes. \u003cem\u003eB. eggerthii\u003c/em\u003e was uniquely under-represented in CRA samples, whereas \u003cem\u003eB. intestinalis, B. plebeius, B. salyersiae\u003c/em\u003e and \u003cem\u003eB. stercoris\u003c/em\u003e were all over-represented. Both, \u003cem\u003eB. thetaiotaomicron and B. xylanisolvens\u003c/em\u003e were under-represented in CRAA samples. \u003cem\u003eB. fragilis\u003c/em\u003e and \u003cem\u003eB. plebeius\u003c/em\u003e were over-represented in CRC. The family Porphyromonadaceae contained 3 species uniquely over-represented in CRC samples, including; \u003cem\u003eP. asaccharolytica, P. endodontalis\u003c/em\u003e and \u003cem\u003eP. uenonis\u003c/em\u003e. Taxonomic markers associated with CRC samples were strongly enriched for species belonging to the Prevotellaceae, including; \u003cem\u003eP. intermedia, P. nigrescens\u003c/em\u003e and \u003cem\u003ePrevotella\u003c/em\u003e sp CAG 520, all of which displayed increased prevalence compared to healthy controls. \u003cem\u003eP. stercorea\u003c/em\u003e was also increased in relative abundance in CRA and CRC samples. Among the family Tannerellaceae, \u003cem\u003eParabacteroides goldsteinii\u003c/em\u003e and \u003cem\u003eP. gordonii\u003c/em\u003e were both enriched in CRA, whereas \u003cem\u003eP. distasonis\u003c/em\u003e was reduced in CRAA samples. Examination of features within Streptococcaceae, we found biomarkers associated with all three disease classes including; \u003cem\u003eS. thermophilus\u003c/em\u003e (CRA), \u003cem\u003eS. mitis\u003c/em\u003e (CRAA) and \u003cem\u003eS. pasteurianus\u003c/em\u003e and \u003cem\u003eS. salivarius\u003c/em\u003e (CRC). A large number of informative features came from the family Lachnospiraceae including; \u003cem\u003eAnaerostipes hadrus\u003c/em\u003e, \u003cem\u003eDorea longicatena\u003c/em\u003e, \u003cem\u003eRoseburia\u003c/em\u003e sp CAG 309, \u003cem\u003eRoseburia\u003c/em\u003e sp CAG 471 all of which were reduced in CRA samples. \u003cem\u003eBlautia wexlerae\u003c/em\u003e, \u003cem\u003eRuminococcus torques\u003c/em\u003e, \u003cem\u003eCoprococcus catus\u003c/em\u003e were also increased but to a smaller extent in CRC. \u003cem\u003eCoprococcus comes\u003c/em\u003e, \u003cem\u003eDorea formicigenerans\u003c/em\u003e, \u003cem\u003eDorea longicatena\u003c/em\u003e, \u003cem\u003eRoseburia faecis\u003c/em\u003e were all increased in CRAA samples and \u003cem\u003eClostridium bolteae\u003c/em\u003e was reduced. \u003cem\u003eEisenbergiella tayi\u003c/em\u003e and \u003cem\u003eClostridium symbiosum\u003c/em\u003e were enriched in CRC samples, whereas \u003cem\u003eRoseburia intestinalis\u003c/em\u003e and \u003cem\u003eRoseburia\u003c/em\u003e sp CAG 303 were reduced in relative abundance. Within the family Peptostreptococcaceae, \u003cem\u003eP. anaerobius\u003c/em\u003e and \u003cem\u003eP. stomatis\u003c/em\u003e both displayed increased relative abundance and stronger increased prevalence in CRC samples. The family Peptoniphilaceae contained \u003cem\u003eAnaerococcus vaginalis\u003c/em\u003e and \u003cem\u003eParvimonas micra\u003c/em\u003e both associated with increased relative abundance and prevalence in CRC samples. Similarly, the Fusobacteriaceae possessed three species associated with CRC, \u003cem\u003eF. naviforme\u003c/em\u003e, \u003cem\u003eF. nucleatum\u003c/em\u003e and \u003cem\u003eFusobacterium\u003c/em\u003e sp oral taxon 370. These results are indicative of biological significance since related genera and species typically behave in a coherent manner in disease.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe have conducted an in-depth meta-analysis of publicly available microbiome shotgun sequence data of fecal samples derived from healthy control donors and those diagnosed by colonoscopy as CRA, CRAA and CRC. Our work describes a novel ensemble approach using two independent algorithms (FIRE and SIAMCAT) for feature selection. This process generated improved external HO AUC for each disease class compared to the best DR baseline model alone. Perhaps more important than improved accuracy of our approach is the likely benefit of increased generalizability achieved using ensemble approaches that reduce biases introduced from single model AI methods. We noted that microbiota sequence data derived from each disease class generated different unique best models (\u003cb\u003eTable\u0026nbsp;2\u003c/b\u003e). Key finding from our studies was the utility of using both taxonomic and gene features to discriminate healthy controls from those with adenomas and carcinomas. Finally, we noted that as the number of discriminatory features increased, the proportion of useful gene features increased relative to taxonomic features, suggesting that taxonomic biomarkers are more finite compared to gene features.\u003c/p\u003e \u003cp\u003eThe characteristics of the 11 studies used varied substantially, including 9 different countries, a focus on different disease states, sampling procedures, variable cohort size, and number of reads passing qc metrics (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Most studies attempted to balance gender and age within their respective cohorts, however male subjects in general were more prevalent than female. Additional factors such as DNA preparation and sequencing methods varied across studies, and importantly, some studies collected fecal samples after colonoscopy. All of these features are expected to create variability in study outcomes. Despite these confounding factors, the taxonomic features identified in individual studies, while variable, define a consensus finding, at least for CRC where the most high quality studies are available for meta-analysis. Given the known inter-personal variability in microbiota and distinct dietary habits of each participating country, the fact that similar taxa are identified strongly suggests that the selective forces operating in CRC are dominant to diet and other known selective pressures.\u003c/p\u003e \u003cp\u003eWe quantified the variability of the datasets used by comparing b-diversity by PCoA (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) and computing Euclidean distance from centroid (\u003cb\u003eSupplementary Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e). These analyses illustrated that large and comparable sample-to-sample variability existed within each study as expected. Most revealing was the outcome on study cross-prediction performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). This analysis showed that some studies generated high predictive power for CRC samples for their own study and in many instances performed well in external study prediction. In no case did training on any one study predict CRC uniformly well across all other studies. Failure of a study to predict samples in external studies may be due to a single outlier study. The factors contributing to study quality are variable and challenging to define.\u003c/p\u003e \u003cp\u003eDespite the various shortcomings of meta-analyses, they are useful when optimizing algorithms that feature high generalizability. The focus of this report investigates feature selection methods and feature type (taxonomic and KO) used for modeling. We evaluated two distinct feature selection methods, FIRE and SIAMCAT. The best performing models generated by machine learning algorithms differ in terms of the features deemed to be of highest importance. Ensemble methods seek to make the best use of multiple models and their underlying selected features to establish both improved performance and increased generalizability [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. We developed a workflow that included two distinct feature selection methods, FIRE and SIAMCAT, separately or in combination (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). We used FIRE that aggregated features selected by the top 5 models. The optimal number of models used in ensemble approaches was not investigated in detail. We established the optimal number of taxonomic or gene features for each disease class. Surprisingly, the optimal number of features for CRC, both for taxonomic and gene features was substantially higher than that determined for CRAA and CRA (\u003cb\u003eTable\u0026nbsp;2\u003c/b\u003e). The reason(s) for this are not clear but may reflect that colorectal tumors have the largest effect on colonic microbiota and their encoded functions, thereby generating a larger spectrum of discriminatory biomarkers. For both CRAA and CRA, a relatively small number of KO features (40) was determined to be optimal. We speculate that this may reflect a relative paucity of discriminatory KO and taxonomic biomarkers at earlier stages of disease.\u003c/p\u003e \u003cp\u003eWe used SIAMCAT (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) as an independent feature selection method that is based on statistical significance of features that we set at (p\u0026thinsp;=\u0026thinsp;\u0026lt;\u0026thinsp;0.001). While FIRE feature selection generally outperformed SIAMCAT, the value of SIAMCAT is evident from cases where the best performance was obtained by combining FIRE and SIAMCAT (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFurthermore, for all analyses involving non-redundant features derived from FIRE and SIAMCAT, SIAMCAT features were always present among the most important features positively contributing to external HO test AUC. Applying SIAMCAT to control and CRC samples generated a taxonomic feature importance list that is highly consistent with taxa reported by several studies [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e], although the features identified through our process are more numerous than generally reported in other studies. CRC taxonomic features are unique as they are enriched for over-represented taxa and species normally resident in the oral cavity. The majority of studies examining these microbes have focused on their behavior in the oral cavity rather than the gut, but accumulating evidence suggests that these taxa are pathobionts capable of causing or contributing to disease in various contexts [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePerhaps most significant is the fact that among the top 24 CRC taxonomic features, all are oral bacteria. These organisms exist in the healthy gut microbiota, but are distinguished from CRC by their increase in relative abundance (fold-change) and an increase in prevalence. Examination of the top 50 CRC features, show that all but 3 taxonomic features display increased relative abundance in CRC relative to control healthy subjects. Among the esxceptions, two features, \u003cem\u003eRoseburia intestinalis\u003c/em\u003e and members of the genus \u003cem\u003eAnaerostipes\u003c/em\u003e are part of the normal commensal gut microbiota. The third case is unexpected involving \u003cem\u003eStreptococcus salivarius\u003c/em\u003e, a known oral bacterium. The reason(s) for the decreased relative abundance of these taxa is unclear.\u003c/p\u003e \u003cp\u003eThe over-representation of oral microbes in CRC fecal samples is consistent with the idea that the tumor microenvironment co-selects these oral species through an unknown fitness advantage that is lacking in healthy individuals and/or a defense mechanism that becomes disabled in CRC. While the factors driving this fitness advantage may be complex, one factor that may explain these results is the metabolic shift occurring in colonic carcinoma epithelium over the transition from health and adenomas to carcinoma. This metabolic shift includes the replacement of oxygen consumption resulting from oxidative metabolism of butyrate for energy to non-oxygen consuming fermentation of lactate, referred to as aerobic glycolysis [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. One important result of this metabolic shift is increased colonic oxygen tension at the site of the tumor. This increased oxygen content may be sufficient to positively select for the aerotolerant oral species observed.\u003c/p\u003e \u003cp\u003eFor all disease classes, the best performance on external test set AUC was obtained using a combination of taxonomic and KO features (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). Upon examining the most important features for each disease class, we observed that external test set AUC for CRC was predominately driven by taxonomic features when a smaller number of features (\u0026lt;\u0026thinsp;50) is used, whereas the most important features for CRAA and CRA were biased for KO features (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). We analyzed the overlap of features across disease classes and surprisingly found that when considering the direction of change, CRA and CRC features overlapping CRAA were very minimal and indeed primarily went in opposite directions, whereas a substantial number of features shared between CRA and CRC were altered in the same direction (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). This result is most surprising and difficult to explain but suggests that the microenvironment and selective pressure of the gut is similar in CRA and CRC but diverges in CRAA. We hypothesize that the higher proportion of shared gene features relative to taxonomic features may reflect the functional redundancy of related and even distantly related taxa that by virtue of shared genes encoded in their respective genomes are essentially inter-changeable within the community.\u003c/p\u003e \u003cp\u003eThese results suggest two surprising conclusions concerning the selective microenvironments generated by colonic lesions and/or the factors produced by gut microbiota that drive disease onset and progression. First, the strong dissimilarity between CRA and CRAA microbiota suggests that advanced adenomas may not simply be larger forms of adenomas. Our results suggest instead indicate that the microbiota adenoma/advanced adenoma interactions represent distinct processes. Second, and perhaps even more surprising, is that given the strong opposing behavior of CRA and CRAA microbiota and gene representation, the CRA and CRC microbiota are more functionally synonymous. Among the 48 KO features that were differentially represented in all 3 disease classes compared to healthy controls, 41 (85%) displayed this pattern of agreement in direction between CRA and CRC and disagreement (15%) in direction between CRAA and the other disease classes. In this regard, the same gene features that are increased in common between CRA and CRC samples are decreased relative to healthy control samples in CRAA and \u003cem\u003evice-versa.\u003c/em\u003e This result is considered preliminary since nearly all of the advanced adenoma samples were derived from a single study. Additional samples will be required to determine whether these patterns of gene representation are generalizable.\u003c/p\u003e \u003cp\u003e \u003cb\u003eConclusions and outlook.\u003c/b\u003e We have developed an ensemble method for feature selection that improves external test set performance of fecal microbiota AUC measures. We noted that meta-analysis of available datasets was associated with several potential confounders including study methods, read depth, and country of origin. Despite confounders and their limitations for meta-analysis, such data represent a powerful resource for developing high quality models that are generalizable. The implementation of FIRE and SIAMCAT displayed positive effects on diagnostic performance, although the best outcomes differed according to disease class, where FIRE alone sometimes outperformed features selected by FIRE and SIAMCAT. The top features generated by FIRE and SIAMCAT combined for CRC are highly consistent with multiple reports analyzing differential representation of taxa in CRC compared to healthy controls. Our results suggest that no single model or set of models are optimal for maximal external test set AUC for all three disease classes that instead require case by case optimization. Our results suggest that the development of an accurate, non-invasive diagnostic test using stool samples to identify all stages of CRC development is highly feasible and that the gut microbiome provides biomarkers that may outperform those used in other commercial CRC tests.\u003c/p\u003e \u003cp\u003eFuture studies should report pathology findings at higher resolution to address open questions such as whether sessile serrated polyps are associated with the same microbes as those harbored in subjects with hyperplastic or tubular adenomas. Additional studies featuring broadened geography and multiple studies from single countries will allow investigators to address whether biomarker variability is a country or study-specific effect. There is a relative paucity of studies examining fecal microbiota in subjects with CRA and CRAA. Additional studies are required to strengthen the diagnostic features associated with these disease states.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eAI\u0026nbsp;artificial intelligence\u003c/p\u003e\n\u003cp\u003eAUC area under the curve\u003c/p\u003e\n\u003cp\u003eCPM\u0026nbsp;counts per million\u003c/p\u003e\n\u003cp\u003eCRA colorectal adenoma\u003c/p\u003e\n\u003cp\u003eCRAA colorectal advanced adenoma\u003c/p\u003e\n\u003cp\u003eCRC colorectal carcinoma\u003c/p\u003e\n\u003cp\u003eDR data robot\u003c/p\u003e\n\u003cp\u003eETBF\u0026nbsp;enterotoxin-producing \u003cem\u003eBacteroides fragilis\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eFIRE\u0026nbsp;feature\u0026nbsp;importance\u0026nbsp;rank\u0026nbsp;ensembling\u003c/p\u003e\n\u003cp\u003eFIT\u0026nbsp;Fecal Immunochemical Test\u003c/p\u003e\n\u003cp\u003eHO hold out\u003c/p\u003e\n\u003cp\u003eKO\u0026nbsp;KEGG ortholog\u003c/p\u003e\n\u003cp\u003eML machine learning\u003c/p\u003e\n\u003cp\u003eNGS\u0026nbsp;next generation sequencing\u003c/p\u003e\n\u003cp\u003ePCoA Principal coordinate analysis\u003c/p\u003e\n\u003cp\u003eSIAMCAT\u0026nbsp;statistical inference of associations between microbial communities and phenotypes\u003c/p\u003e\n\u003cp\u003eSNM\u0026nbsp;supervised normalization method\u003c/p\u003e\n\u003cp\u003eTMM trimmed mean of M-values\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and Consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors have given their consent for the publication.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study used data already available in public databases:\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eFeng et al. 2015\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cbr\u003e\u0026nbsp;https://www.ncbi.nlm.nih.gov/bioproject/PRJEB7774\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eGao et al. 2021\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003ehttps://www.ncbi.nlm.nih.gov/bioproject/PRJNA514108\u003c/p\u003e\n \u003cp\u003ehttps://www.ncbi.nlm.nih.gov/bioproject/PRJNA706060\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eGupta et al 2019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cbr\u003e\u0026nbsp;https://www.ncbi.nlm.nih.gov/bioproject/PRJNA531273\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eHanningan et al. 2018\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003ehttps://www.ncbi.nlm.nih.gov/bioproject/PRJNA389927\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eThomas et al. 2019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cbr\u003e\u0026nbsp;https://www.ncbi.nlm.nih.gov/bioproject/PRJNA447983\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eVogtmann et al. 2016\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003ehttps://www.ncbi.nlm.nih.gov/bioproject/PRJEB12449\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eWirbel et al. 2016\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cbr\u003e\u0026nbsp;https://www.ncbi.nlm.nih.gov/bioproject/PRJEB27928\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eYachida et al. 2019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003ehttps://www.ncbi.nlm.nih.gov/bioproject/PRJDB4176\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eSpanoggianopoulis et al. 2022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003ehttps://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA720145\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eYu et al. 2017\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003ehttps://www.ncbi.nlm.nih.gov/bioproject/PRJEB10878\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003eZeller et al. 2014\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"50%\" valign=\"top\"\u003e\n \u003cp\u003ehttps://www.ncbi.nlm.nih.gov/bioproject/PRJEB6070\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eS.N.P., A.E, P.K., G.K., E.F., Y.V., and T.K. are all employees of Prescient Metabiomics LLC, who are developing CRC stool diagnostic test. Other authors have no conflicts to report.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was supported by private investor funding to PMB\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; Contribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eS.N.P. wrote the manuscript and conducted data analysis. A.M.E, P.Z.K., E.F., N.S., F.F., Y.V., and G.K. conceived and conducted data analysis. T.K. conceived and directed the study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements.\u0026nbsp;\u003c/strong\u003eThe authors would like to thank Dan Knights and Emily Hollister for helpful discussion and critique of the manuscript and the staff at Diversigen for technical assistance. This work was supported by private funding provided to Prescient Metabiomics.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018;68:394\u0026ndash;424. \u003c/li\u003e\n\u003cli\u003eLauby-Secretan B, Vilahur N, Bianchini F, Guha N, Straif K. The IARC Perspective on Colorectal Cancer Screening. N Engl J Med. 2018;378:1734\u0026ndash;40. \u003c/li\u003e\n\u003cli\u003eLee JK, Liles EG, Bent S, Levin TR, Corley DA. Accuracy of Fecal Immunochemical Tests for Colorectal Cancer: Systematic Review and Meta-analysis. Ann Intern Med. 2014;160:171\u0026ndash;81. \u003c/li\u003e\n\u003cli\u003eHundt S, Haug U, Brenner H. Comparative evaluation of immunochemical fecal occult blood tests for colorectal adenoma detection. Ann Intern Med. 2009;150:162\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eImperiale TF, Ransohoff DF, Itzkowitz SH, Levin TR, Lavin P, Lidgard GP, et al. Multitarget Stool DNA Testing for Colorectal-Cancer Screening. N Engl J Med. 2014;370:1287\u0026ndash;97. \u003c/li\u003e\n\u003cli\u003eFrank C, Sundquist J, Yu H, Hemminki A, Hemminki K. Concordant and discordant familial cancer: Familial risks, proportions and population impact: Concordant and discordant familial cancer. Int J Cancer. 2017;140:1510\u0026ndash;6. \u003c/li\u003e\n\u003cli\u003eFrancescangeli F, De Angelis ML, Zeuner A. Dietary Factors in the Control of Gut Homeostasis, Intestinal Stem Cells, and Colorectal Cancer. Nutrients. 2019;11:2936. \u003c/li\u003e\n\u003cli\u003eJohnson CM, Wei C, Ensor JE, Smolenski DJ, Amos CI, Levin B, et al. Meta-analyses of colorectal cancer risk factors. Cancer Causes Control. 2013;24:1207\u0026ndash;22. \u003c/li\u003e\n\u003cli\u003eHuxley RR, Ansary-Moghaddam A, Clifton P, Czernichow S, Parr CL, Woodward M. The impact of dietary and lifestyle risk factors on risk of colorectal cancer: A quantitative overview of the epidemiological evidence. Int J Cancer. 2009;125:171\u0026ndash;80. \u003c/li\u003e\n\u003cli\u003eGarcia S, Pruitt SL, Singal AG, Murphy CC. Colorectal cancer incidence among Hispanics and non-Hispanic Whites in the United States. Cancer Causes Control. 2018;29:1039\u0026ndash;46. \u003c/li\u003e\n\u003cli\u003eGreathouse KL, White JR, Padgett RN, Perrotta BG, Jenkins GD, Chia N, et al. Gut microbiome meta-analysis reveals dysbiosis is independent of body mass index in predicting risk of obesity-associated CRC. BMJ Open Gastroenterol. 2019;6:e000247. \u003c/li\u003e\n\u003cli\u003eLee SH, Yun Y, Kim SJ, Lee E-J, Chang Y, Ryu S, et al. Association between Cigarette Smoking Status and Composition of Gut Microbiota: Population-Based Cross-Sectional Study. J Clin Med. 2018;7:E282. \u003c/li\u003e\n\u003cli\u003eRodriguez-Gonzalez A, Orio L. Microbiota and Alcohol Use Disorder: Are Psychobiotics a Novel Therapeutic Strategy? Curr Pharm Des. 2020;26:2426\u0026ndash;37. \u003c/li\u003e\n\u003cli\u003eAllen JM, Mailing LJ, Niemiro GM, Moore R, Cook MD, White BA, et al. Exercise Alters Gut Microbiota Composition and Function in Lean and Obese Humans. Med Sci Sports Exerc. 2018;50:747\u0026ndash;57. \u003c/li\u003e\n\u003cli\u003eDai Z, Coker OO, Nakatsu G, Wu WKK, Zhao L, Chen Z, et al. Multi-cohort analysis of colorectal cancer metagenome identified altered bacteria across populations and universal bacterial markers. Microbiome. 2018;6:70. \u003c/li\u003e\n\u003cli\u003eThomas AM, Manghi P, Asnicar F, Pasolli E, Armanini F, Zolfo M, et al. Metagenomic analysis of colorectal cancer datasets identifies cross-cohort microbial diagnostic signatures and a link with choline degradation. Nat Med. 2019;25:667\u0026ndash;78. \u003c/li\u003e\n\u003cli\u003eWirbel J, Pyl PT, Kartal E, Zych K, Kashani A, Milanese A, et al. Meta-analysis of fecal metagenomes reveals global microbial signatures that are specific for colorectal cancer. Nat Med. 2019;25:679. \u003c/li\u003e\n\u003cli\u003eArthur JC, Perez-Chanona E, M\u0026uuml;hlbauer M, Tomkovich S, Uronis JM, Fan T-J, et al. Intestinal Inflammation Targets Cancer-Inducing Activity of the Microbiota. Science. 2012;338:120\u0026ndash;3. \u003c/li\u003e\n\u003cli\u003eKostic AD, Chun E, Robertson L, Glickman JN, Gallini CA, Michaud M, et al. Fusobacterium nucleatum Potentiates Intestinal Tumorigenesis and Modulates the Tumor-Immune Microenvironment. Cell Host Microbe. 2013;14:207\u0026ndash;15. \u003c/li\u003e\n\u003cli\u003eWu S, Rhee K-J, Albesiano E, Rabizadeh S, Wu X, Yen H-R, et al. A human colonic commensal promotes colon tumorigenesis via activation of T helper type 17 T cell responses. Nat Med. 2009;15:1016\u0026ndash;22. \u003c/li\u003e\n\u003cli\u003eRobinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinforma Oxf Engl. 2010;26:139\u0026ndash;40. \u003c/li\u003e\n\u003cli\u003eLaw CW, Chen Y, Shi W, Smyth GK. voom: Precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol. 2014;15:R29. \u003c/li\u003e\n\u003cli\u003ePoore GD, Kopylova E, Zhu Q, Carpenter C, Fraraccio S, Wandro S, et al. Microbiome analyses of blood and tissues suggest cancer diagnostic approach. Nature. NIH Public Access; 2020;579:567. \u003c/li\u003e\n\u003cli\u003eMecham BH, Nelson PS, Storey JD. Supervised normalization of microarrays. Bioinforma Oxf Engl. 2010;26:1308\u0026ndash;15. \u003c/li\u003e\n\u003cli\u003eWirbel J, Zych K, Essex M, Karcher N, Kartal E, Salazar G, et al. Microbiome meta-analysis and cross-disease comparison enabled by the SIAMCAT machine learning toolbox. Genome Biol. 2021;22:93. \u003c/li\u003e\n\u003cli\u003eYu J, Feng Q, Wong SH, Zhang D, Liang Q yi, Qin Y, et al. Metagenomic analysis of faecal microbiome as a tool towards targeted non-invasive biomarkers for colorectal cancer. Gut. BMJ Publishing Group; 2017;66:70\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eFeng Q, Liang S, Jia H, Stadlmayr A, Tang L, Lan Z, et al. Gut microbiome development along the colorectal adenoma\u0026ndash;carcinoma sequence. Nat Commun. 2015;6:6528. \u003c/li\u003e\n\u003cli\u003eZeller G, Tap J, Voigt AY, Sunagawa S, Kultima JR, Costea PI, et al. Potential of fecal microbiota for early-stage detection of colorectal cancer. Mol Syst Biol. 2014;10:766. \u003c/li\u003e\n\u003cli\u003eYachida S, Mizutani S, Shiroma H, Shiba S, Nakajima T, Sakamoto T, et al. Metagenomic and metabolomic analyses reveal distinct stage-specific phenotypes of the gut microbiota in colorectal cancer. Nat Med. 2019;25:968\u0026ndash;76. \u003c/li\u003e\n\u003cli\u003eGao R, Zhu Y, Kong C, Xia K, Li H, Zhu Y, et al. Alterations, Interactions, and Diagnostic Potential of Gut Bacteria and Viruses in Colorectal Cancer. Front Cell Infect Microbiol. 2021;11:657867. \u003c/li\u003e\n\u003cli\u003eGupta A, Dhakan DB, Maji A, Saxena R, P.K. VP, Mahajan S, et al. Association of Flavonifractor plautii, a Flavonoid-Degrading Bacterium, with the Gut Microbiome of Colorectal Cancer Patients in India. Bordenstein S, editor. mSystems. 2019;4:e00438-19, /msystems/4/6/msys.00438-19.atom. \u003c/li\u003e\n\u003cli\u003eVogtmann E, Hua X, Zeller G, Sunagawa S, Voigt AY, Hercog R, et al. Colorectal Cancer and the Human Gut Microbiome: Reproducibility with Whole-Genome Shotgun Sequencing. PLOS ONE. 2016;11:e0155362. \u003c/li\u003e\n\u003cli\u003eSpanogiannopoulos P, Kyaw TS, Guthrie BGH, Bradley PH, Lee JV, Melamed J, et al. Host and gut bacteria share metabolic pathways for anti-cancer drug metabolism. Nat Microbiol. 2022;7:1605\u0026ndash;20. \u003c/li\u003e\n\u003cli\u003eHannigan GD, Duhaime MB, Ruffin MT, Koumpouras CC, Schloss PD. Diagnostic Potential and Interactive Dynamics of the Colorectal Cancer Virome. mBio. 2018;9:e02248-18. \u003c/li\u003e\n\u003cli\u003eBeghini F, McIver LJ, Blanco-M\u0026iacute;guez A, Dubois L, Asnicar F, Maharjan S, et al. Integrating taxonomic, functional, and strain-level profiling of diverse microbial communities with bioBakery 3. eLife [Internet]. eLife Sciences Publications, Ltd; 2021 [cited 2022 Mar 31];10. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8096432/\u003c/li\u003e\n\u003cli\u003eKanehisa M, Furumichi M, Tanabe M, Sato Y, Morishima K. KEGG: new perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res. 2017;45:D353\u0026ndash;61. \u003c/li\u003e\n\u003cli\u003eBose S, Das C, Banerjee A, Ghosh K, Chattopadhyay M, Chattopadhyay S, et al. An ensemble machine learning model based on multiple filtering and supervised attribute clustering algorithm for classifying cancer samples. PeerJ Comput Sci. 2021;7:e671. \u003c/li\u003e\n\u003cli\u003eDrewes JL, White JR, Dejea CM, Fathi P, Iyadorai T, Vadivelu J, et al. High-resolution bacterial 16S rRNA gene profile meta-analysis and biofilm status reveal common colorectal cancer consortia. NPJ Biofilms Microbiomes. 2017;3:34. \u003c/li\u003e\n\u003cli\u003ePushalkar S, Ji X, Li Y, Estilo C, Yegnanarayana R, Singh B, et al. Comparison of oral microbiota in tumor and non-tumor tissues of patients with oral squamous cell carcinoma. BMC Microbiol. 2012;12:144. \u003c/li\u003e\n\u003cli\u003eLoftus M. Bacterial community structure alterations within the colorectal cancer gut microbiome. 2021;18. \u003c/li\u003e\n\u003cli\u003eWang G, Yu Y, Wang Y-Z, Wang J-J, Guan R, Sun Y, et al. Role of SCFAs in gut microbiome and glycolysis for colorectal cancer therapy. J Cell Physiol. 2019;234:17023\u0026ndash;49. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"colorectal cancer, stool microbiome, artificial intelligence, machine learning, biomarkers","lastPublishedDoi":"10.21203/rs.3.rs-2838129/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2838129/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDespite the effectiveness of colonoscopy for reducing colorectal cancer (CRC) mortality, poor screening compliance ranks CRC as the second most deadly malignancy. There is a need to develop a preventative, non-invasive diagnostic test, such as a fecal microbiota test, for early detection of both pre-cancerous adenomas and carcinomas to effectively reduce mortality.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe conducted a clinical meta-analysis of published deep metagenomic stool sequence datasets including 1,670 subjects from 9 countries, including 703 healthy controls, 161 precancerous colorectal adenoma (CRA), 48 advanced precancerous colorectal adenoma (CRAA) and 758 CRC cases diagnosed by colonoscopy. We analyzed these data through a novel automated machine learning workflow using a two-stage feature importance ranking and ensemble modeling method to identify and select highly predictive taxonomic and functional biomarkers. Machine learning modeling of selected features differentiated the metagenomic profiles of healthy patients from CRA, CRAA and CRC cases with an average area under the curve (AUC) for external holdout testing of 0.84 (sensitivity=0.82; specificity=0.71, accuracy=0.77) for CRC; an AUC of 0.97 (sensitivity=0.78; specificity=0.98, accuracy=0.97) for CRAA; and an AUC of 0.90 (sensitivity=0.74, specificity=0.89, accuracy=0.86) for CRA. These performance outcomes represented a 2%, 3% and 8% increase in AUC, compared to baseline ML performance, respectively. The predictive features identified for each disease class were largely distinct and represented differing proportions of taxonomic and functional features.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe predictive taxonomic features identified for each disease class were largely distinct, whereas many functional gene features were shared across disease classes but displayed differing direction of change. Application of our ensemble approach for feature selection increased the predictive power of each disease class and moreover may generate discriminatory models with greater generalizability.\u003c/p\u003e","manuscriptTitle":"Meta-analysis of Multi-functional Biomarkers for Discovery and Predictive Modeling of Colorectal Adenoma and Carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-04-25 14:05:17","doi":"10.21203/rs.3.rs-2838129/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d635d9a7-0e50-49df-a345-004d3cd13ecb","owner":[],"postedDate":"April 25th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2023-05-08T17:29:16+00:00","versionOfRecord":[],"versionCreatedAt":"2023-04-25 14:05:17","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-2838129","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2838129","identity":"rs-2838129","version":["v1"]},"buildId":"WrCJVZZCHTDjtuVLN7oU0","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}