Introduction
Antibiotic resistance is one of the most urgent threats in modern medicine. As more bacteria become resistant to existing drugs, scientists need faster and more reliable ways to discover and improve new treatments.
In our new paper published in Nature Machine Intelligence, we introduce ApexGO, an artificial intelligence system designed to help create stronger antibiotic candidates more efficiently.
What is ApexGO?
ApexGO is a generative AI platform built to optimize antimicrobial peptides: small protein-like molecules that can kill bacteria.
Unlike approaches that only search through existing libraries of molecules, ApexGO begins with promising peptide templates and suggests sequence changes that may improve their antibacterial activity. In simple terms, it acts as a smart design engine. It learns patterns from peptide sequences, explores new molecular possibilities, and proposes improved versions that can then be synthesized and tested in the lab.
From AI design to laboratory testing
To evaluate ApexGO, the team started with 10 peptide templates and used the system to design optimized derivatives. From these AI-designed molecules, the researchers chemically synthesized 100 candidates.
The compounds were then tested in detail to understand:
- How well they killed bacteria.
- How they acted against microbial cells.
- What structures they formed.
- Whether they showed signs of toxicity to host cells.
The results were especially promising against Gram-negative bacteria, a group of difficult-to-treat pathogens responsible for many serious and hospital-acquired infections.
Strong experimental results
ApexGO achieved an 85 percent experimental hit rate, meaning that most of the AI-designed molecules showed real antimicrobial activity when tested in the lab.
The system also improved antimicrobial activity in 72 percent of cases against clinically relevant Gram-negative pathogens. This suggests that ApexGO can do more than identify possible antibiotics: it can help improve molecules that already show promise.
Testing in infection models
The researchers also tested selected AI-optimized molecules in mouse models of Acinetobacter baumannii infection. A. baumannii is a dangerous Gram-negative bacterium known for its resistance to many existing drugs.
In these studies, some ApexGO-designed molecules outperformed the original peptide templates they were based on. Their activity was comparable to, or better than, that of a last-resort antibiotic.
Conclusion
This work points to a future in which generative AI becomes a practical part of drug discovery. By helping researchers move from promising starting molecules to experimentally validated candidates more quickly, tools like ApexGO could make a meaningful difference in the global fight against drug-resistant infections.
For antibiotic research, the shift is significant: AI is moving beyond the discovery of promising molecules and becoming a tool for actively engineering stronger therapeutic candidates.
For more information on this study, please refer to the full paper published in Nature Machine Intelligence: https://www.nature.com/articles/s42256-026-01237-5
For more information, please contact:
Machine Biology Group
University of Pennsylvania
Authors:
Marcelo D. T. Torres, Yimeng Zeng, Fangping Wan, Natalie Maus, Jacob Gardner & Cesar de la Fuente-Nunez
Published:
May 13, 2026
About Machine Biology Group:
The mission statement of the Machine Biology Group at the University of Pennsylvania is to use the power of machines to accelerate discoveries in biology and medicine.