Unveiling the evolution of antimicrobial peptides in gut microbes via foundation-model-powered framework

Abstract

Abstract Antimicrobial resistance poses a growing threat to public health, emphasizing the urgent need for novel therapeutic strategies. Antimicrobial peptides (AMPs), short peptide sequences with diverse mechanisms of action, offer a promising alternative due to their broad-spectrum activity against pathogens. Recent advances in protein language models (PLMs) have revolutionized protein structure prediction and functional annotation, highlighting their potential for AMP discovery and therapeutic development. In this context, we present AMP-SEMiner (Antimicrobial Peptide Structural Evolution Miner), an AI-driven framework designed to identify AMPs from metagenome-assembled genomes (MAGs). By integrating PLMs, structural clustering and evolutionary analysis into the framework, AMP-SEMiner can identify AMPs encoded by small open reading frames (smORFs) and encrypted peptides (EPs), significantly expanding the discovery space. Using this approach, we identified 1,670,600 AMP candidates from diverse habitats. Experimental validation of 29 candidates revealed antimicrobial activity in 18, with 13 surpassing antibiotics in effectiveness. Further analysis of AMPs from human gut microbiomes demonstrated both conserved and adaptive evolutionary strategies, ensuring their functional efficacy in the dynamic gut environment. These findings position AMP-SEMiner as a powerful tool for the discovery and characterization of novel AMPs, with significant potential to drive the development of new antimicrobial therapies.