Abstract
The human microbiome is increasingly mined for diagnostic and therapeutic biomarkers using machine learning (ML). However, metagenomics-specific software is scarce, and overoptimistic evaluation and limited cross-study generalization are prevailing issues. To address these, we developed SIAMCAT, a versatile R toolbox for ML-based comparative metagenomics. We demonstrate its capabilities in a meta-analysis of fecal metagenomic studies (10,803 samples). When naively transferred across studies, ML models lost accuracy and disease specificity, which could however be resolved by a novel training set augmentation strategy. This reveals some biomarkers to be disease-specific, with others shared across multiple conditions. SIAMCAT is freely available from siamcat.embl.de.
Highlights
The study of microbial communities through metagenomic sequencing has begun to uncover how communities are shaped by—and interact with—their environment, including the host organism in the case of gut microbes [1, 2]
Machine learning and statistical analysis workflows implemented in SIAMCAT The SIAMCAT R package is a versatile toolbox for analyzing microbiome data from case-control studies
When comparing taxonomic and functional profiles derived from the same dataset, we found a high correlation between AUROC values (Pearson’s r = 0.92, P < 2 × 10−16), on average taxonomic profiles performed slightly better than functional profiles (Additional file 1: Figure S7)
Summary
The study of microbial communities through metagenomic sequencing has begun to uncover how communities are shaped by—and interact with—their environment, including the host organism in the case of gut microbes [1, 2]. As the microbiome is increasingly recognized as an important factor in health and disease, many possibilities for clinical applications are emerging for diagnosis [8, 9], prognosis, or prevention of disease [10]. The prospect of clinical applications comes with an urgent need for methodological rigor in microbiome analyses in order to ensure the robustness of findings. It is necessary to assess the clinical value of biomarkers identified from the microbiome in an unbiased manner— by their statistical significance, but more importantly by their prediction accuracy on independent samples Additional issues arise from key characteristics of metagenomic data such as large technical and inter-individual variation [12], experimental bias [13], compositionality of relative abundances, zero inflation, and non-Gaussian distribution, all of which necessitate data normalization in order for ML algorithms to work well
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
