Overview
- UPenn researchers detailed a method in Nature Machine Intelligence that tunes existing antimicrobial peptides rather than screening vast libraries.
- The approach starts from a promising peptide, proposes precise edits, and guides the next round using model predictions.
- The team combined their earlier APEX predictor with Bayesian optimization to choose which variants to test next.
- In lab and mouse studies, 85% of designs blocked bacterial growth, 72% beat their parent peptides, and two matched polymyxin B in mice.
- The authors said these are early leads that need work on safety, stability, and how long they act before any move toward patients.