A McMaster AI explored 46 billion molecules and designed a new antibiotic that cleared drug-resistant staph in mice

Researchers at McMaster University in Canada reported that a generative artificial-intelligence model they built, called SyntheMol-RL, designed a structurally new antibiotic that controlled a drug-resistant infection in animal tests. The work, announced on April 23, 2026, is part of a wider effort to use AI to refill a pipeline of new antibiotics that has run dangerously dry even as bacteria keep evolving resistance to existing drugs.

The scale of the search is what makes the approach powerful. Rather than screening libraries of chemicals that already exist, SyntheMol-RL assembles molecules from about 150,000 building blocks using a set of 50 reliable chemical reactions, letting it explore a theoretical space of up to 46 billion possible compounds, far beyond what any laboratory could physically test. Crucially, the updated model was trained to favor candidates that are water-soluble and practical to synthesize, so the molecules it proposes can actually be made and absorbed by the body.

“The work, announced on April 23, 2026, is part of a wider effort to use AI to refill a pipeline of new antibiotics that has run dangerously dry even as bacteria keep evolving resistance to existing drugs.”

From a test batch of 79 AI-proposed antibacterials, the team identified a promising new compound they named synthecin, which effectively controlled drug-resistant Staphylococcus aureus wound infections in mouse models. The project was led by assistant professor Jon Stokes with graduate students Gary Liu and Denise Catacutan. As Stokes's group put it, the system "configures those fragments in different ways, faster than humans ever could, to create new, larger chemical compounds."

The honest caveats are significant. Synthecin's results come from cell and mouse studies, not people, and the team has not yet pinned down exactly how it kills bacteria, a step needed to establish its safety profile before any human trials. Most promising molecules at this stage still fail along the long road to an approved medicine. Even so, a tool that designs genuinely novel, makeable antibiotics, rather than tweaking old ones, gives a stalled field a fresh and badly needed way forward.