AI-driven discovery of hypoglycemic quorum sensing peptides from the human gut microbiome

The human gut microbiome encodes a large spectrum of quorum sensing peptides (QSPs) playing a vital role in microbial communication and host-microbial interactions, yet discovery and functional characterization of novel QSPs are still limited. Here, we developed an integrated artificial intelligence (AI)-driven pipeline combining large-scale genome mining, machine learning-based screening, Alphafold3 structural modelling, and molecular dynamics simulations to systematically identify 1,670 novel candidate QSPs (cQSPs) from 12,488 human gut microbial genomes. Subsequent experimental screening identified 29 cQSPs with α-glucosidase inhibitory activities. Notably, GIVLVPTSLVSG (GIVL), an Agr-derived QSP from Clostridium tyrobutyricum, not only exhibited canonical quorum-sensing functions, but also demonstrated a stable interaction with and strong inhibitory activity against α-glucosidase. In diabetic mice, administration of GIVL significantly improved glucose homeostasis and insulin sensitivity accompanied by increased production of short-chain fatty acids (SCFAs) in the gut microbiome. Our findings expand the QSP database and establish a generalizable, AI-enabled framework for the systematic discovery of microbiome-derived peptides with therapeutic potential in metabolic disease, positioning QSPs as promising sources of therapeutics for diabetes and other chronic metabolic disorders.