Artificial Intelligence Discovers Antibiotic in Record Time

AI antibiotic
MIT researchers used a machine-learning algorithm to identify a drug called halicin that kills many strains of bacteria. Halicin (top row) prevented the development of antibiotic resistance in E. coli, while ciprofloxacin (bottom row) did not. Image: Courtesy of the Collins Lab at MIT (CC BY-NC-ND 3.0)

In 1928, a Scottish scientist named Sir Alexander Fleming left his lab where he was studying the staphylococcus bacteria to go on a two-week vacation with his family. When he returned to his lab bench, he not only realized he hadn't tidied his work space very well, but that the dishes with the bacteria in them were growing mold. He also noticed that the bacteria seemed to be actively avoiding the moldy areas of the petri dish. Later he said "I certainly didn't plan to revolutionize all medicine by discovering the world's first antibiotic, or bacteria killer. But I suppose that was exactly what I did."

These days it doesn't take a slovenly scientist to discover important new antibiotics — it just takes a computer. A group of researchers at the Massachusetts Institute of Technology (MIT) have used artificial intelligence (AI) to identify a new antibiotic that kills even some hitherto antibiotic-resistant strains.


But does this mean they staffed the lab with robots rather than people? Nope! The research team created a computer model that systematically screened more than a hundred million chemical compounds in just a few days — a feat that would take lab technicians many years (and a lot of the same sort of scientific serendipity that visited Fleming) to accomplish.

Very few new antibiotics have been discovered in the past decade, during which time bacteria are getting tougher.

"We're facing a growing crisis around antibiotic resistance, and this situation is being generated by both an increasing number of pathogens becoming resistant to existing antibiotics, and an anemic pipeline in the biotech and pharmaceutical industries for new antibiotics," said James Collins, a professor in MIT's Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering, in a press release.


How They Did It

The research team developed a machine-learning computer model that could identify about 2,500 molecular compounds that prohibited the growth of bacteria — in this case, E. coli, specifically. They then introduced the program to 6,000 drugs that are currently being studied to see if any of them might be useful in curing known human diseases. Once the model selected the molecule with the strongest antibacterial potential that didn't look similar to any known antibiotics, the team used a different model to see if the molecule would be detrimental to people.

Et voila! The model narrowed the candidates down to one — the researchers have dubbed it "halicin" — which has been tested in the past as a drug to treat diabetes. Halicin has been tested on lab samples of several different antibiotic-resistant strains of bacteria and has been shown to kill almost all of them, with the exception of one very stubborn lung pathogen.


After discovering halicin, the research team used the model to identify 23 more candidates using another database of compounds and found two that were particularly powerful. The researchers are now working to find antibiotics that are more selective in the bacteria they kill, so they don't destroy all our beneficial gut flora while saving our lives. As for halicin, the researchers plan to work with a pharmaceutical company or nonprofit to develop the drug for use in humans, according to the press release.