Antibiotic discovery is notoriously hard and slow. What if we could change that by optimizing antibiotics on the computer? Introducing APEXgo, our latest AI model, which integrates a transformer-based variational autoencoder with Bayesian optimization to design novel antibiotics. Unlike traditional AI approaches, which rely on fixed databases of known molecules, APEXgo explores uncharted sequence space, proposing entirely novel peptides that have not been seen before. We put APEXgo to the test with a bold challenge: could it design optimized compounds with antibiotic activity using templates from extinct organisms like woolly mammoths and giant sloths? 🐘🦣 The results? 🤯 · 100 peptides were synthesized and characterized for antimicrobial activity, mechanism of action, secondary structure, and cytotoxicity. · APEXgo achieved an 85% experimental hit rate. · 72% success in improving activity against dangerous Gram-negative pathogens —a major challenge. · The antibiotics designed by APEXgo were effective in mouse models, even outperforming polymyxin B (a last-resort antibiotic). This work represents the first ground-truth experimental validation of generative Bayesian optimization in any setting, and we are excited that its debut application is in antibiotic discovery. With APEXgo, we have entered an era where AI accelerates antibiotic discovery and explores molecular diversity at digital speed. #AI #Antibiotics Huge congratulations to my amazing Team and collaborators on this achievement: @mdt_torres, Yimeng Zeng, Fangping Wan, Natalie Maus, and Jake Gardner @Penn @PennMedicine @PennEngineers Link to the paper: biorxiv.org/content/10.1101/…