Unravelling metabolite-microbiome interactions in inflammatory bowel disease through AI and interaction-based modelling

Inflammatory Bowel Diseases (IBDs) are chronic inflammatory disorders of the gastrointestinal tract and colon affecting approximately 7 million individuals worldwide. The knowledge of specific pathology and aetiological mechanisms leading to IBD is limited, however a reduced immune system, antibiotic use and reserved diet may initiate symptoms. Dysbiosis of the gut microbiome, and consequently a varied composition of the metabolome, has been extensively linked to these risk factors and IBD. Metagenomic sequencing and liquid-chromatography mass spectrometry (LC-MS) of N = 220 fecal samples by Fransoza et al., provided abundance data on microbial genera and metabolites for use in this study. Identification of differentially abundant microbes and metabolites was performed using a Wilcoxon test, followed by feature selection of random forest (RF), gradient-boosting (XGBoost) and least absolute shrinkage operator (LASSO) models. The performance of these features was then validated using RF models on the Human Microbiome Project 2 (HMP2) dataset and a microbial community (MICOM) model was utilised to predict and interpret the interactions between key microbes and metabolites. The Flavronifractor genus and microbes of the families Lachnospiraceae and Oscillospiraceae were found differential by all models. Metabolic pathways commonly influenced by such microbes in IBD were CoA biosynthesis, bile acid metabolism and amino acid production and degradation. This study highlights distinct interactive microbiome and metabolome profiles within IBD and the highly potential pathways causing disease pathology. It therefore paves way for future investigation into new therapeutic targets and non-invasive diagnostic tools for IBD. • The study analyzed faecal samples using metagenomic sequencing and LC-MS data sets • We used AI and machine learning models that includes identifying microbes and metabolites through statistical and machine learning models using random forest (RF), gradient-boosting (XGBoost) and least absolute shrinkage operator (LASSO) models • The performance of these features was then validated on the Human Microbiome Project 2 • We found that the Flavronifractor genus and microbes from the Lachnospiraceae and Oscillospiraceae families were consistently linked to IBD, impacting key metabolic pathways like CoA biosynthesis and bile acid metabolism. • Microbial community modeling identified seven metabolites were recognized as significantly differentially produced or consumed by the microbial communities of the IBD and control samples