Researchers at the University of Pennsylvania reported ApexGO, an AI-powered method designed to improve peptide antibiotic candidates starting from a small set of imperfect molecules and iterating modifications guided by a predictive algorithm. In reported results, 85% of AI-generated peptides halted bacterial growth in lab tests, and 72% outperformed the peptides from which they were derived. In mice, two ApexGO-designed antimicrobial peptides reduced bacterial counts to levels comparable to the antibiotic polymyxin B. The work, published in Nature Machine Intelligence, positions ApexGO as a more directed search strategy across chemical space, with the authors describing real-world predictive performance rather than purely model-to-model optimization.