Particle.news

Download on the App Store

Generative AI Designs New Antimicrobial Peptides With Drug-Level Efficacy in Mice

The Penn-built diffusion model uses Meta’s ESM-2 to generate biologically plausible sequences prioritized by an internal ranker for synthesis.

Overview

  • A Cell Biomaterials study unveils AMP-Diffusion, a University of Pennsylvania tool that creates antimicrobial peptide candidates de novo.
  • The system generated roughly 50,000 sequences, with APEX 1.1 narrowing the set to 46 that were synthesized for laboratory and animal tests.
  • In mouse skin-infection models, two peptides matched the efficacy of levofloxacin and polymyxin B with no adverse effects reported.
  • Across assays, 76% of tested peptides killed bacteria, including multidrug-resistant strains, while showing low toxicity.
  • Researchers describe the work as proof of principle and aim to steer designs toward drug-like properties as they address translation to real-world therapies.