Antibiotic resistance is one of the most serious threats to global health. Experts estimate that by 2050, drug-resistant bacteria could kill up to 10 million people annually. The World Health Organization has placed Acinetobacter baumannii on its list of the most dangerous resistant pathogens: it causes pneumonia, meningitis, and wound infections, and is extremely difficult to treat. In March 2024, researchers from McMaster University and Stanford introduced a system that could become part of the answer.

How SyntheMol Works

The system is called SyntheMol — short for “synthesizing molecules.” It is a generative AI model that doesn’t simply search for suitable molecules in existing databases but designs entirely new ones. “We wanted to use AI to design completely new molecules that have never been seen in nature,” explains James Zou, one of the study’s authors.

Technically, it works like a construction set. The model has a library of 132,000 molecular fragments — essentially “Lego pieces” with different chemical properties. It combines them using 13 validated chemical reactions, generating around 30 billion possible molecular variants. For comparison: even large-scale laboratory screenings cover at most about a million compounds. AI works with a chemical space that traditional methods simply cannot reach.

The Synthesis Recipe — The Key Innovation

The main problem with previous attempts to use AI for drug design was that the modeled molecules were often impossible to synthesize in the lab. Chemists would get a beautiful structure on screen — but without a clear way to build it.

SyntheMol solves this. The model produces not only the molecule’s structure but also a detailed “recipe” for its synthesis — a step-by-step pathway from the starting fragments to the finished compound. “Generating such recipes is a new approach and a game changer, because chemists don’t know how to make AI-designed molecules,” says James Zou.

Results

In just nine hours, the model generated around 25,000 potential antibiotics. The researchers selected the 70 most promising compounds and sent them to the Ukrainian chemical company Enamine for synthesis. Of the 70, 58 were successfully synthesized, and six showed potent antibacterial activity against A. baumannii. Moreover, these compounds were also effective against other dangerous bacteria, including E. coli, Klebsiella pneumoniae, and MRSA — methicillin-resistant Staphylococcus aureus.

Additionally, every molecule was run through a separate AI model trained to predict toxicity. Two of the six compounds were tested in mice and found to be safe. The other four did not dissolve in water, which requires further formulation work.

Importantly, all six molecules are structurally different from each other and from existing antibiotics. This reduces the risk that bacteria will quickly develop resistance to them.

Why This Matters Right Now

“Antibiotics are a unique medicine. As soon as we start using them in the clinic, a countdown begins until they become ineffective, because bacteria evolve quickly to resist them,” says Jonathan Stokes, the study’s lead author. A steady stream of new drugs is needed — and they must be found quickly and inexpensively. This is where AI demonstrates its core value: it radically accelerates and reduces the cost of the process.

In 2026, the team presented an updated version — SyntheMol-RL — which works with a space of 46 billion possible compounds, uses 150,000 building blocks and 50 reactions. The new version has already produced an antibiotic called synthecin, which proved effective against drug-resistant staph infections in mouse models.

The researchers are also expanding the model’s applications — from heart disease to fluorescent molecules for laboratory research. SyntheMol shows that AI can do more than accelerate science: it can open up regions of chemical space that humans have never explored.

Powered by Froala Editor