Penn researchers built a GenAI model that designs novel antibiotics, showing early success against drug-resistant bacteria in animal tests.
Researchers at the University of Pennsylvania (Penn) have unveiled AMP-Diffusion, a GenAI model that designs new antimicrobial peptides (AMPs). Unlike traditional discovery methods limited by natural datasets, this tool creates antibiotic candidates from scratch. The goal is to tackle the growing crisis of antimicrobial resistance, which continues to outpace conventional drug development.
AMP-Diffusion adapts diffusion models—commonly used in image generation—to craft amino acid sequences. It builds on Meta’s protein language model, ESM-2, which provides structural insights into proteins. By combining this foundation with generative design, the system produced over 50,000 peptide sequences. A secondary AI, APEX 1.1, then filtered and ranked the best candidates for synthesis.
Out of 46 lab-tested peptides, several matched the effectiveness of FDA-approved drugs like levofloxacin and polymyxin B in animal models. Importantly, these AI-designed molecules showed no detectable side effects. This marks a major step forward, as it demonstrates that GenAI can not only search for but actually invent drug candidates with real therapeutic potential.
The researchers aim to refine AMP-Diffusion to design peptides tailored for specific pathogens or with built-in drug-like properties. Their long-term vision is to compress the antibiotic discovery process from years to days. For now, the results highlight how GenAI can accelerate drug development, expand the pipeline of antibiotic candidates, and help counter one of the greatest challenges in modern medicine.
