UP chemists develop AI tool to predict proteins that can be used as antibiotics

The University of the Philippines Diliman - College of Science (UPD-CS) has created an artificial intelligence tool that can accelerate the discovery of new antibacterial peptides, which are proteins that can potentially be used as antibiotics.

Developed by Remmer Salas, Dr. Portia Mahal Sabido, and Dr. Ricky Nellas of the UPD-CS Institute of Chemistry, the AI tool—named ISCAPE or Interpretable Support Vector Classifier of Antibacterial Activity of Peptides against Escherichia coli—helps researchers predict whether a peptide can kill or inhibit the growth of E. coli bacteria.

E. coli can cause illness such as food poisoning, diarrhea, and urinary tract infections.

Unlike traditional processes that require synthesizing and testing antibacterial peptides one by one, the system only needs a Simplified Molecular Input Line-Entry System (SMILES) string to evaluate candidate molecules.

"We used AI to learn from existing data and identify patterns that distinguish active peptides from inactive ones," Salas said.

ISCAPE also shows which molecular features make a peptide effective. This helps researchers save time and resources as it reduces trial-and-error experiments and allows them to design better peptides more efficiently.

Early-stage screening

However, Salas clarified that it does not replace laboratory experiments and only makes discovery more efficient by helping researchers focus on the most promising candidates.

"ISCAPE helps address antimicrobial resistance by accelerating early-stage screening through data-driven peptide design," Salas said.

The AI-powered model can be adapted to predict activity against bacteria beyond E. coli, but it must first be retrained using high-quality, strain-specific datasets.

It could also be applied to predict the activity of other bioactive peptides.

"Applying ISCAPE to other biological targets requires well-curated datasets with experimentally validated activity," Salas said.

"The model must then be retrained using the molecular features we identified as optimal for peptides," Salas added.

ISCAPE is publicly available for other scientists. The web server can be accessed through Hugging Face Spaces. The training data and code for large-scale prediction are available on GitHub.

The research titled "Interpretable support vector classifier for reliable prediction of antibacterial activity of modified peptides against Escherichia coli" can be found in the Journal of Molecular Graphics and Modelling. — VDV, GMA News