In a new breakthrough, MIT researchers have used generative AI algorithms to design antibiotic candidates capable of killing two of the world’s most stubborn bacterial threats: drug-resistant Neisseria gonorrhoeae and multi-drug-resistant Staphylococcus aureus (MRSA).

The team generated more than 36 million hypothetical compounds — molecules that do not currently exist — and computationally screened them for antimicrobial properties. The best-performing candidates were structurally distinct from all existing antibiotics and appear to act by new mechanisms that disrupt bacterial cell membranes — something that would have been impossible through human trial and error.

“We’re excited about the new possibilities that this project opens up for antibiotics development. Our work shows the power of AI from a drug design standpoint, and enables us to exploit much larger chemical spaces that were previously inaccessible,” says James Collins, senior author of the study and the Termeer Professor of Medical Engineering and Science in MIT’s Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering.

Antibiotic resistance is a growing global crisis. Drug-resistant bacterial infections are estimated to cause nearly 5 million deaths per year worldwide, while most new antibiotics approved in recent decades are simply variations on older ones.

By venturing into “underexplored” chemical space, AI enables researchers to identify compounds with mechanisms the bacteria have never encountered — making resistance less likely to emerge quickly.

The researchers began by using generative AI to design molecules around a specific chemical fragment already known to have antimicrobial activity, targeting N. gonorrhoeae. They compiled a library of 45 million known chemical fragments, combining basic atoms and chemical groups from public and commercial databases.

Machine-learning models trained on antibacterial activity predicted which fragments were most promising, narrowing the list from 45 million to about 1 million after filtering out toxic, unstable, or antibiotic-like candidates.

“We wanted to get rid of anything that would look like an existing antibiotic, to help address the antimicrobial resistance crisis in a fundamentally different way. By venturing into underexplored areas of chemical space, our goal was to uncover novel mechanisms of action,” says Aarti Krishnan, an MIT postdoctoral researcher and lead author of the study.

One fragment, F1, emerged as a strong lead. Using two algorithms — CReM (Chemically Reasonable Mutations, which creates new molecules by adding, replacing, or removing atoms and chemical groups) and F-VAE (fragment-based variational autoencoder, which builds complete molecules around a fragment based on patterns it has learned from large chemical databases) — they generated about 7 million F1-based molecules.

From these, NG1 stood out: it killed drug-resistant gonorrhea in lab dishes and mouse models, and works by targeting LptA, a protein involved in building the bacterial outer membrane. The drug appears to kill the bacteria by disrupting the process they use to build their outer membrane — an essential structure they can’t survive without.

Next, the team applied unconstrained generative AI to target S. aureus, including MRSA. Here, the algorithms had no fragment requirement — they simply followed chemical plausibility rules to generate over 29 million molecules.

After similar filtering, about 90 candidates remained. Of the 22 molecules synthesized for testing, six showed strong antibacterial activity and were effective against MRSA in the lab. The lead candidate, DN1, cleared MRSA skin infections in mice and works by broadly disrupting bacterial membranes, with effects that aren’t limited to targeting a single protein.

Nonprofit Phare Bio, part of the Antibiotics-AI Project, is now refining NG1 and DN1 through medicinal chemistry to make them suitable for further preclinical testing.

Collins said the team is also exploring AI-designed compounds for other high-priority pathogens, including Mycobacterium tuberculosis and Pseudomonas aeruginosa.

Q: What did MIT researchers achieve using AI?
A: They used generative AI to design entirely new antibiotic candidates for drug-resistant bacteria, including MRSA and gonorrhea.

Q: How many compounds did AI evaluate?
A: Over 36 million theoretical molecules were generated and screened computationally.

Q: How are these antibiotics different from existing ones?
A: They are structurally distinct and appear to work by novel mechanisms, making them harder for bacteria to resist.

Q: What are NG1 and DN1?
A: NG1 targets N. gonorrhoeae by interfering with membrane synthesis, while DN1 kills MRSA by disrupting its cell membrane.

Q: Who is advancing the drug candidates toward testing?
A: The nonprofit Phare Bio is refining the compounds for preclinical development.

This work shows how generative AI can dramatically accelerate drug discovery by searching chemical spaces far beyond existing compound libraries. Similar approaches — like DeepMind’s AlphaFold protein structure predictions and AI-driven drug design at other biotech companies — are expanding the possibilities for creating more effective treatments.

For global health, advances like these could help close the gap between the rapid rise of antibiotic resistance and the much slower pace of traditional drug development, giving researchers a way to design new medicines before dangerous pathogens gain the upper hand.

For the AI field, it’s a clear example of machine learning moving beyond prediction into creation — generating not just ideas, but physical, testable solutions. As these AI-designed compounds advance toward clinical testing, they could mark a turning point in how we fight some of the deadliest pathogens.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.