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

Antimicrobial resistance poses a major threat to public health, prompting the development of alternative therapies such as antimicrobial peptides (AMPs). Protein language models (PLMs) have advanced protein structure and function predictions, facilitating AMP discovery. We developed antimicrobial peptide structural evolution miner (AMP-SEMiner), an AI-driven framework that integrates PLMs, structural clustering, and evolutionary analysis to systematically identify AMPs encoded by small open reading frames and AMP-containing proteins in metagenome-assembled genomes. AMP-SEMiner identified over 1.6 million AMP candidates across diverse environments. Experimental validation showed antimicrobial activity in 9 of the 20 tested candidates, with 5 surpassing antibiotic effectiveness; variant peptides derived from these candidates similarly demonstrated strong antimicrobial efficacy. AMPs from human gut microbiomes revealed both conserved and adaptive evolutionary strategies, reflecting their dynamic ecological roles. AMP-SEMiner thus represents a valuable tool for expanding AMP discovery and has significant potential to inform the development of alternative antimicrobial treatments.