AI found a drug that kills one superbug only

The World Health Organization lists it as a top priority for new antibiotics. Researchers at MIT and McMaster University trained a neural network on how roughly 7,500 known molecules affected the bacterium's growth, then asked it to screen thousands more it had never tested.

The model flagged a compound the team named abaucin. In lab testing it stopped A. baumannii specifically, leaving other bacteria alone, and it controlled an infected wound in mice. Narrow-spectrum activity like that is unusual and valuable, since it is less likely to breed resistance across unrelated species.

Abaucin's mechanism, disrupting how the bacterium moves proteins to its outer membrane, was unknown before the model pointed to the compound. It has not yet been tested in human trials, and turning a promising mouse result into an approved drug is a years-long process. The same neural network approach has since been pointed at other resistant bacteria.

A machine cut a key for one lock.