Understanding IBS with XAI

Abstract

Irritable Bowel Syndrome (IBS) is a prevalent gastrointestinal disorder with a complex etiology, where alterations in the gut microbiome and metabolome are increasingly recognised as playing a significant role. This study aimed to identify robust IBS-associated microbial and metabolic signatures and explore their interactions through a comprehensive multi-omics analysis of the PRJNA812699 cohort, coupled with cross-cohort validation. We analysed 16S rRNA metagenomics, metatranscriptomics, and metabolomics data individually and in an integrated multi-omics dataset using Random Forest classifiers for IBS prediction and feature importance analysis to identify key biomarkers. Explainable AI (XAI) methods, including feature relationship analysis via Random Forest Regressor and network graph visualisation, were employed to understand the complex interactions between these features. Our analysis of the primary cohort revealed that metabolomics data exhibited strong predictive power for IBS, and multi-omics integration significantly enhanced classification accuracy. Cross-cohort validation across three independent IBS cohorts and a larger cross-cohort dataset identified recurring bacterial taxa, including Prevotella copri and genera like Bacteriodes and Streptococcus, as potential generalizable IBS markers. Feature relationship analysis highlighted key interactions, such as the central role of Odoribacter splanchnicus at the species level and Streptococcus at the genus level in the cross-cohort network. These findings provide a multifaceted view of the IBS gut microbiome, underscoring the value of multi-omics integration and cross-cohort validation in identifying robust biomarkers and understanding the intricate biological networks underlying IBS pathophysiology. This research contributes to the growing body of knowledge on IBS, offering potential avenues for future diagnostic and therapeutic development.