Particle.news

Download on the App Store

Penn Team Debuts AMP‑Diffusion, an AI That Designs Antibiotic Peptides Effective in Mice

The Cell Biomaterials study validates de novo antibiotic design using a diffusion model guided by ESM‑2 with an APEX filter.

Overview

  • AMP‑Diffusion generated roughly 50,000 antimicrobial peptide candidates, which were ranked and narrowed to 46 for synthesis using the APEX 1.1 filter.
  • Laboratory testing showed 76% of the synthesized peptides inhibited bacteria, including multidrug‑resistant strains, with low toxicity.
  • Two AI‑designed peptides treated skin infections in mice with efficacy comparable to levofloxacin and polymyxin B, without observable adverse effects.
  • The system integrates Meta’s ESM‑2 protein language model during generation to produce biologically plausible sequences more efficiently.
  • Authors describe a proof‑of‑concept and plan to steer designs toward specific pathogens and drug‑like properties, with funding sources and potential conflicts disclosed.