Discovery of antimicrobial peptides in the global microbiome with machine learning

A machine-learning-based approach predicts antimicrobial peptides (AMPs) within the global microbiome using a vast dataset of 63,410 metagenomes and 87,920 prokaryotic genomes. The AMPSphere catalog contains 863,498 non-redundant peptides, providing insights into their evolutionary origins. The approach validates predictions by synthesizing and testing 100 AMPs against drug-resistant pathogens and human gut commensals. 79 active AMPs were identified, with 63 targeting pathogens and disrupting bacterial membranes. This approach provides an open-access resource for antibiotic discovery.