Learning from networks: leveraging metabolic models and machine learning to discover microbial metabolic functions and interactions

Most microbes live in communities and shape their environments through metabolic interactions, yet linking individual microorganisms and genes to community-level functions remains challenging. We integrate computational and experimental approaches to investigate microbial metabolism across scales, from individual species to complex microbiomes. By combining network-based analyses, genome-scale metabolic modelling, and machine learning, we infer microbial interactions from metagenomes, predict metabolic capabilities, and propagate functional information from metagenomes to individual species. These predictions generate experimentally testable hypotheses, enabling the discovery of previously uncharacterised metabolic functions, genes, and interactions, elucidating microbe-environment relationships across habitats, from the human gut to the oceans.