The world sorely needs new antibiotics. Can A.I. help?

The global health community has long warned of a possible superbug catastrophe.

Last year, the Centers for Disease Control (CDC) updated the list of potential antibiotic-resistant threats to include up to 18 bacteria and fungi. These are the kinds of pathogens that countries are ill prepared to tackle in the midst of an epidemic (bugs such as the current coronavirus aren’t exactly in the same camp, though they do present their own unique threats). There are more than 2.8 million cases of antibiotic-resistant superbug infections in the U.S. each year, and more than 35,000 deaths, according to the CDC.

But could machine learning and artificial intelligence fuel the search for new antibiotics? MIT researchers believe so.

A so-called “deep learning” algorithm was deployed by MIT scientists to uncover new antibiotics that use different mechanisms from the ones currently on the market. And the scientists say that, using this technology, they were able to identify a new type of antibiotic they’ve dubbed halicin that can kill off stubborn bacterial strains that may be resistant to current treatments, including Clostridium difficile, Acinetobacter baumannii, and Mycobacterium tuberculosis.

“Our approach revealed this amazing molecule which is arguably one of the more powerful antibiotics that has been discovered,” said James Collins, the Termeer Professor of Medical Engineering and Science in MIT’s Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering and one of the study’s senior authors, in a statement. The paper’s first author is Jonathan Stokes of MIT and the Broad Institute of MIT and Harvard.

Regina Barzilay, a professor of computer science at MIT and another one of the study’s senior authors, compared the process of using machine learning for antibiotic drug discovery to the way that Amazon curates its recommendations to consumers.

“What machine learning and A.I. can help with is, it can make the process faster and cheaper,” she told Fortune. “The way that people discover molecules in lab involves lots of experimentation. You need to manufacture molecules, try them out in the cell. The process involves a lot of trial and error.”

By contrast, a team of researchers—with an assist from these algorithmic helpers—can program the right kind of A.I. to sniff out promising drug candidates without having to go through as expensive of a clinical song-and-dance.

That’s important since there’s a fundamental buzzsaw to funding antibiotic and preventive medical research, given that these kinds of treatments may not bring in as much money as therapies which treat diseases after the fact.

“A chemist knows there’s a certain combination of atoms more likely to kill a bug because that’s what their experience has been,” said Barzilay. “Machines can try to identify the combination between the molecular structure and how a combination of things can kill certain bacteria.”

MIT claims that this machine learning model can also help identify other promising antibiotic candidates. The details are preliminary, as tends to happen with early-stage science, but the underlying effort focuses on using A.I. to home in on chemical drugs that are different from what’s already on the market in order to fight hard-to-control bacteria.

The World Health Organization (WHO) says that at least 12 antibiotic-resistant superbugs pose the greatest threat to the global health population—and that doesn’t even include all of the bacteria and fungi on the CDC’s list of resistant pathogens in the U.S.

Antibiotics aren’t the only field seeing a boom in A.I.-assisted drug discovery. As pharmaceutical companies’ return on investment for drug R&D plummets, big pharma companies and startup biotechs alike are turning to machine learning for an efficiency boost.