In a groundbreaking study that combines artificial intelligence and laboratory science, researchers have discovered a new antibiotic for inflammatory bowel disease (IBD) after an AI model accurately predicted the drug's effectiveness before the team could experimentally prove it.
The study, published on October 3, 2025, in the journal Nature Microbiology, promises the fast-track development of more innovative and safer antibiotics.
The new drug, called enterololin, is a narrow-spectrum antibiotic intended to target a group of bacteria known as the Enterobacteriaceae, which includes adherent-invasive E. coli commonly found in the gut of patients with Crohn's disease and ulcerative colitis.
In contrast to the broad-spectrum antibiotics that indiscriminately target disease-inducing microbes and others, with effects resembling a nuclear bomb on the gut flora, enterololin selectively targets disease-inducing microbes while sparing beneficial microbes, thereby decreasing the chances of drug resistance and dysbiosis.
The role played by AI in explaining the mechanism of action of the drug is what makes this discovery unprecedented.
Following the discovery of enterololin, scientists at McMaster University collaborated with the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), utilising an AI model called DiffDock, to analyse the compound.
The system could predict within 100 seconds that the working compound was enterololin, which interfered with the LolCDE complex- a protein complex needed by bacteria to survive.
"The AI provided the researchers with a narrow, falsifiable hypothesis, which would otherwise take months and millions of dollars to identify," according to Jon Stokes, assistant professor at McMaster and the primary investigator on the study. It did not supplant the science, but it is a powerful push in that direction.
Graduate student Denise Catacutan, under the guidance of Stokes, then spent six months confirming the AI's prediction by performing conventional laboratory experiments—a process that typically takes up to two years.
These findings confirmed that, in fact, the LolCDE complex is inhibited by enterololin, and the lab-evolved resistant mutants hybridised at the same loci as the genes that the AI was flagging.
The drug has already demonstrated potential in mouse models, repressing AIEC infection without causing a significant disturbance to the overall microbiome.
Enterololin exhibits low mammalian cytotoxicity and a low rate of resistance, making it an attractive candidate for clinical development.
The compound has been licensed to Stokes Bio, a spin-out company established to develop the discovery. Only in three years would human trials be possible.
“This isn’t just about finding one new drug,” said Stokes.
“It’s about proving a new model for drug discovery—where AI doesn’t just guess molecules, but helps us understand them,” he added.