Researchers have used machine learning to fish two promising new antibiotics out of a chemical haystack, both of them active against drug-resistant gonorrhea and at least one working in a way no existing treatment does. The work, from a team led by James Collins at Harvard's Wyss Institute, MIT and the Broad Institute, was published last month in Science Translational Medicine and offers a rare bit of good news in a field defined by dwindling options.
The bacterium behind gonorrhea, Neisseria gonorrhoeae, has become one of public health's clearest warnings about antibiotic resistance, having evolved to defeat nearly every drug deployed against it and leaving clinicians reliant on essentially a single frontline therapy. Finding replacements is hard: promising compounds are scarce, and resistance to a new drug typically emerges within five to ten years of its introduction.
The team approached the search at scale. It physically tested 38,650 small molecules for activity against the bacterium, then used machine-learning models to evaluate roughly six million more compounds virtually, looking for chemical structures that killed the pathogen without resembling drugs already in use. That funnel yielded 213 validated candidates, which the researchers narrowed to two standouts they labeled MP20 and A1.
A1 is the more intriguing of the pair. An aminothiazole compound, it kills the bacterium by binding and inhibiting alanine racemase, an enzyme the microbe needs to construct its cell wall — a target no current gonorrhea antibiotic exploits. Because resistance is usually specific to a drug's mechanism, a compound attacking an entirely new target is less likely to be undercut by the resistance the bacterium has already evolved.
The findings are early. Both compounds have been tested in the laboratory and in animal models, not in people, and the long road from a promising molecule to an approved drug winnows out most candidates along the way. What the study demonstrates is less a finished medicine than a method: using machine learning to widen the search for antibiotics far beyond what brute-force lab screening can cover, and to prioritize the rare hits worth pursuing.
That method may be the more durable result. Gonorrhea infects tens of millions of people a year worldwide, with more than 600,000 cases annually in the United States alone, and its resistance profile makes it a bellwether for other pathogens moving the same direction. A discovery pipeline that can keep surfacing new mechanisms, rather than variations on old ones, is the thing the field has most lacked.