Abstract
The rise of antibiotic resistance has created an urgent need for the discovery of new antimicrobial peptides (AMPs), prompting various screening strategies. However, the mechanisms of action of AMPs are often overlooked during screening and optimization. Here, a mechanism-driven screening approach is introduced using machine learning-based computational models to identify peptide sequences that target bacterial membranes and form pores. From the metaproteomes of poison frogs, African clawed frogs, and human skin, seven peptides are identified and validated, each exhibiting antimicrobial activity with minimal hemolysis and cytotoxicity. These peptides demonstrated membrane disruption in liposome leakage assays, with three showing broad-spectrum activity against Gram-positive and Gram-negative bacteria. Single-molecule experiments confirmed peptide oligomerization on membranes, while electrophysiological measurements verified pore formation by the three broad-spectrum AMPs, suggesting a correlation between pore-forming ability and broad-spectrum antimicrobial activity. This approach offers a promising mechanism-driven strategy for discovering new antimicrobial agents to combat antibiotic resistance.