Antibiotic resistance represents a global challenge, driven by the rise of multidrug-resistant pathogens such as carbapenem-resistant Acinetobacter baumannii. Antimicrobial peptides (AMPs) offer a promising alternative to conventional antibiotics due to their broad-spectrum activity, rapid bactericidal mechanisms, and reduced tendency to induce resistance. To accelerate AMPs discovery, a pre-trained protein large language model called ProteoGPT was developed and then refined through transfer learning with specialized tools for AMP classification, cytotoxicity prediction, and peptide generation. This framework enables the screening of hundreds of millions of natural peptides and the generation of novel sequences, many of which displayed potent antimicrobial activity. The selected AMPs demonstrated strong efficacy in vitro against both sensitive and resistant bacterial strains, as well as antifungal activity, while maintaining low toxicity. In vivo mouse infection models confirmed a therapeutic performance comparable to clinical antibiotics, with limited organ damage and minimal disruption of gut microbiota.
This approach highlights how generative artificial intelligence can expand the accessible peptide space and accelerate the development of next-generation antimicrobials against superbugs.
Suggested by PD Dr. Raquel Mejías Luque