Evolutionary Design of Membrane-Lytic Antimicrobial Peptides with Mixture of Experts

AMPainterV2 evolved membrane-lytic antimicrobial peptides in vitro, with the best candidate reaching a mean minimal inhibitory concentration (MIC) of 1.5 M and stronger activity than polymyxin B.

• Why it matters: Existing AMP design models do not explicitly model membrane disruption, limiting mechanism-driven optimization against drug-resistant pathogens.
• What they did: AMPainterV2 uses generative adversarial imitation learning with a Mixture-of-Experts policy to perform insertion, deletion, and mutation during evolutionary design.
• The result: All ten top-ranked peptides evolved from random sequences were active membrane-lytic AMPs, and a deletion-only variant produced six miniaturized peptides with an average 36% length reduction and comparable or improved activity.

AI summary of the abstract · may contain errors · not medical advice · read the paper.