Machine-learning screen finds new antibiotic candidates for drug-resistant gonorrhoea
A team led by researchers at MIT, with collaborators at Sweden’s Karolinska Institutet, used a deep-learning model to hunt for antibiotics against Neisseria gonorrhoeae, the bacterium that causes gonorrhoea and is increasingly resistant to existing drugs. As reported by Nature, the model was trained on tens of thousands of experimentally tested molecules and then used to screen roughly six million compounds, surfacing 83 candidates with confirmed activity against the pathogen.
Two molecules, designated MP20 and A1, stood out because their chemical structures do not resemble antibiotics already in use — a property that matters because resistance often spreads across drugs that share a mechanism. In mice and an organ-on-a-chip model of infection, both compounds killed the bacteria quickly and selectively without inducing resistance. According to the underlying study in Science Translational Medicine, MP20 disrupts the bacterial membrane and damages DNA, while A1 targets an enzyme needed to build the cell wall.
The result is a method advance as much as a drug one: it shows machine-learning screens can find genuinely new chemical scaffolds rather than variations on known antibiotics. The candidates remain preclinical, and the long, often-failed path from animal models to approved human drugs still lies ahead.