Deep learning reveals functional archetypes in the adult human gut microbiome that underlie interindividual variability and confound disease signals

Abstract

Understanding the functional diversity of the gut microbiome is essential for decoding its roles in health and disease. Using a deep-learning framework, we identified three functional archetypes defining healthy adult gut microbiomes, each characterized by specific metabolic potentials: sugar metabolism with branched-chain amino acid and cell wall synthesis (Archetype 1), fatty acid and TCA cycle metabolism (Archetype 2), and amino acid and nitrogen metabolism (Archetype 3). Archetype proximity is linked to stability, with Archetype 2 representing the most resilient state, likely due to its metabolic flexibility. Functional diversity emerged as a confounder in disease-associated microbial signatures. In inflammatory bowel disease, we observed archetype-specific shifts, including increased carbohydrate metabolism in Archetype 1-dominant samples and inflammatory pathways in Archetype 3-dominant samples, suggesting distinct opportunities for microbiome-targeted interventions. This framework addresses key challenges in microbiome research, including inter-individual variability and confounding, while providing robust insights into disease-associated functional shifts and microbial ecosystem dynamics. HighlightsO_LIAdult gut microbiomes are defined by three functional archetypes C_LIO_LIArchetypes reveal distinct metabolic potentials and inform on microbiome stability C_LIO_LIArchetype-specific functional profiles confound disease associations and reveal therapeutic targets C_LIO_LIA deep-learning framework enables robust characterization of microbial functional ecosystems C_LI