MIT uses generative AI to design new antibiotic candidates against drug-resistant bacteria

Researchers at MIT reported in the journal Cell that they had used generative artificial intelligence to design entirely new antibiotic candidates, then synthesized and tested a handful against two dangerous, drug-resistant bacteria. The work, led by James Collins of MIT's Antibiotics-AI Project, targeted Neisseria gonorrhoeae, which causes drug-resistant gonorrhea, and methicillin-resistant Staphylococcus aureus, known as MRSA.

Rather than screening libraries of existing chemicals, the team used algorithms to generate more than 36 million theoretical compounds and computationally filtered them for antimicrobial promise and low predicted toxicity. The most interesting candidates were structurally distinct from known antibiotics and appeared to work through novel mechanisms that disrupt bacterial cell membranes. One compound, NG1, targets a protein called LptA involved in building the bacterial outer membrane.

“The work, led by James Collins of MIT's Antibiotics-AI Project, targeted Neisseria gonorrhoeae, which causes drug-resistant gonorrhea, and methicillin-resistant Staphylococcus aureus, known as MRSA.”

The researchers synthesized a small subset and found two standouts. NG1 was effective against drug-resistant gonorrhea in laboratory and mouse models, while a compound named DN1 cleared MRSA skin infections in mice. "We are excited about the new possibilities that this project opens up for antibiotics development," Collins said.

Important caveats apply. These are early-stage results in cell cultures and animals, not humans, and the long road from a promising molecule to an approved drug includes safety, dosing and manufacturing hurdles that most candidates do not survive. The team plans to advance leads through Phare Bio, a nonprofit Collins helped create. Still, with antimicrobial resistance contributing to well over a million deaths a year, a method that designs structurally new candidates rather than tweaking old ones offers a genuinely useful new path for a field that has seen few breakthroughs in decades.