MIT's AI researchers find a new antibiotic in an old drug

Researchers at MIT used a machine-learning algorithm to uncover the potent antibiotic properties hiding within a small-molecule drug that had been explored as a potential diabetes treatment.

In early mouse models and lab tests, the compound was able to kill off some of the most drug-resistant and dangerous bacteria using a unique mechanism of action compared to other antibiotics.

“We wanted to develop a platform that would allow us to harness the power of artificial intelligence to usher in a new age of antibiotic drug discovery,” said James Collins, MIT’s Termeer Professor of Medical Engineering and Science.

“Our approach revealed this amazing molecule which is arguably one of the more powerful antibiotics that has been discovered,” said Collins, who served as a co-senior author of a paper published in the journal Cell.

And, in the face of growing antibiotic resistance, Collins pointed to the need for new screening methods that could help revitalize the industry’s pipeline for developing new antibiotics, a costly process that has seen few successes in recent decades compared to other pharmaceutical research fields.

While other antibiotics may target the structure of bacterial cell walls or interfere with an infection’s enzyme or protein synthesis activity, this new molecule aims to block the electrochemical gradient that flows across cell membranes and is necessary for maintaining cellular metabolism.

Dubbed halicin—after the thinking machine in "2001: A Space Odyssey"—the compound was shown to be incredibly effective against strains of Clostridium difficile, Acinetobacter baumannii and Mycobacterium tuberculosis. Additionally, according to researchers, its new mechanism of action could potentially sidestep most bacterial methods of drug resistance.

“When you’re dealing with a molecule that likely associates with membrane components, a cell can’t necessarily acquire a single mutation or a couple of mutations to change the chemistry of the outer membrane,” first author Jonathan Stokes, an MIT and Broad Institute postdoc, said to MIT News. “Mutations like that tend to be far more complex to acquire evolutionarily.”

The predictive, in silico model used to identify the compound was first trained on about 2,500 chemical structures know to be effective at killing E. coli. After identifying patterns and commonalities in those data, the model then screened a library of about 6,000 compounds from the Drug Repurposing Hub at the Broad Institute of MIT and Harvard.

One hit was predicted to have strong antibacterial activity, though researchers said they identified several other candidates they plan to test further. A separate machine-learning model was then used to gauge halicin’s toxicity against human cells, before it was synthesized for testing.

In the lab, researchers found that E. coli was unable to develop any resistance to halicin after 30 days, compared to the 24 to 72 hours needed to defend against the antibiotic ciprofloxacin. The team said it plans to study the drug further and work with nonprofit or industry partners to develop it for use in humans.