AI system transforms weak antibiotics into powerful treatments

Researchers at University of Pennsylvania have developed a powerful new artificial intelligence system capable of reforming imperfect antibiotic candidates into stronger and more effective treatments.

The new method, known as ApexGO, uses AI to improve promising molecules step by step rather than by searching large chemical databases for potential antibiotics.

Scientists behind the project say the technology could speed up the discovery of life-saving medicines at a time when antibiotic resistance is becoming a growing worldwide threat.

A new approach to antibiotic discovery

Unlike many AI systems already used in drug discovery, ApexGO does not begin with millions of unknown compounds. Instead, it starts with a small number of partially effective molecules and refines them through a series of calculated modifications.

“Antibiotic discovery is fundamentally a search problem across an enormous molecular space, ApexGO gives us a way to navigate that space with far more direction,” said César de la Fuente, co-senior author of the study. “ApexGO begins with a promising but imperfect peptide, proposes precise edits, predicts whether those changes are likely to enhance antimicrobial activity and then keeps moving toward versions that are more likely to work when we make and test them.”

Unlike many AI systems already used in drug discovery, ApexGO does not begin with millions of unknown compounds

Laboratory testing showed encouraging results. Researchers found that 85 percent of AI-designed molecules successfully stopped bacterial growth, while 72 percent performed better than the original peptides from which they were developed.

In mouse studies, two antimicrobial peptides generated by ApexGO reduced bacterial levels at rates comparable to polymyxin B, an FDA-approved antibiotic used as a last-resort treatment against certain resistant infections.

“What is striking is that ApexGO’s predictions held up in the real world,” said Jacob Gardner, Assistant Professor in Computer and Information Science and co-senior author of the paper. “ApexGO was optimising against another computer model, so one concern was that it might find molecules that looked good to the model but failed in the lab. Instead, the majority of the molecules it designed actually worked.”

Building on earlier AI research

The project builds on earlier work by de la Fuente’s laboratory, which previously developed an AI model called APEX that predicts whether peptides may possess antimicrobial properties.

For years, researchers in the lab have searched for antibiotic candidates in different and unusual places, including frog secretions and ancient microbes.

“APEX helped us find promising antibiotic candidates in enormous biological datasets,” said Marcelo Torres, Research Assistant Professor of Psychiatry in the Perelman School of Medicine and co-first author of the paper. “ApexGO takes the next step: once we have a promising molecule, it helps us ask how to make it better.”

APEX helped us find promising antibiotic candidates in enormous biological datasets

The system relies on Bayesian optimisation, a machine learning approach that helps AI systems efficiently explore large numbers of possibilities without testing every single option individually.

Faster drug discovery

Although the newly designed peptides remain early-stage candidates requiring further testing for safety and stability, scientists believe the technology could significantly reduce the time needed to identify viable drug treatments.

“ApexGO shows that AI can do more than predict which molecules might work: it can help us improve them,” said de la Fuente. “At a time when antibiotic resistance is rising worldwide, we need technologies that help us move faster from an idea to a real therapeutic candidate – ApexGO is an important step toward that future.”