Photo by Jo Lin on Unsplash
MANILA, PHILIPPINES [TAC] — Researchers at the University of the Philippines Diliman have developed an artificial intelligence tool to accelerate the discovery of new antibacterial treatments, offering a potential weapon against the global threat of antimicrobial resistance.
The machine learning tool, named ISCAPE, is designed to screen and predict whether specific small molecules—known as antimicrobial peptides—can effectively target and inhibit the growth of Escherichia coli bacteria.
Public health officials view the development of these peptides as a critical frontier in medicine. Decades of antibiotic overuse have allowed strains of bacteria to mutate, rendering conventional treatments increasingly ineffective against common infections.
“Traditionally, discovering antibacterial peptides means synthesizing many candidates and testing them one by one—a process that is time-consuming,” said Remmer Salas, a researcher at the university’s Institute of Chemistry and one of the tool’s developers. “We used AI to learn from existing data and identify patterns that distinguish active peptides from inactive ones.”
Opening the Black Box
Unlike standard “black box” artificial intelligence models that mask their internal logic, the ISCAPE system is interpretable. It details the specific molecular and chemical features that make a given peptide effective against bacteria.
The system operates via an open-access web server where scientists can input a text-based chemical string to immediately evaluate candidate molecules, bypassing early-stage, trial-and-error laboratory experiments.
“ISCAPE helps address antimicrobial resistance by accelerating early-stage screening through data-driven peptide design,” Salas said, noting that while the software does not replace physical laboratory validation, it allows researchers to focus resources strictly on the most viable chemical candidates.
Scalable Architecture
The development team, which includes chemistry professors Dr. Portia Mahal Sabido and Dr. Ricky Nellas, published their peer-reviewed findings in the Journal of Molecular Graphics and Modelling.
While the initial model was trained explicitly to identify peptides effective against E. coli, the mathematical approach can be scaled to target other pathogens. The researchers stated that the algorithm can be retrained to predict defenses against diverse bacterial strains, provided it is fed high-quality, experimentally validated biological datasets.
To encourage international collaboration, the scientists have made the tool’s underlying code, dataset, and predictive framework publicly available on GitHub and Hugging Face.
