MIT Uses Generative AI To Develop Two Novel Antibiotics Targeting Drug-Resistant Gonorrhea And MRSA

In Brief

MIT researchers used AI to design two novel antibiotics, NG1 and DN1, that successfully target drug-resistant gonorrhea and MRSA in mice, highlighting AI’s potential to transform antibiotic discovery.

Massachusetts Institute of Technology (MIT) researchers have employed AI to develop two novel antibiotics effective against drug-resistant gonorrhea and MRSA, potentially offering new strategies to combat infections responsible for millions of deaths each year

By leveraging generative AI algorithms, the team created over 36 million potential compounds and computationally screened them for antimicrobial activity. The most promising candidates are structurally unique compared to existing antibiotics and appear to act through previously unseen mechanisms that disrupt bacterial cell membranes. This method enabled the generation and evaluation of entirely new compounds, and the researchers plan to extend the approach to design antibiotics targeting other bacterial species

Most new antibiotics approved over the past 45 years are variations of existing drugs, while bacterial resistance continues to rise, causing nearly 5 million deaths annually

In order to tackle this, MIT’s Antibiotics-AI Project employed AI to explore both existing compounds and entirely new, hypothetical molecules. Using machine learning models trained to predict antibacterial activity, the team first screened millions of chemical fragments, eliminating those likely to be toxic or similar to existing antibiotics

They then applied two generative AI algorithms: CReM, which modifies molecules by adding, replacing, or deleting atoms and groups, and F-VAE, which constructs full molecules from fragments based on learned chemical patterns. This AI-driven process generated roughly 7 million candidate molecules, which were computationally screened for activity against N. gonorrhoeae

From this, about 1,000 compounds were shortlisted, 80 were synthetically feasible, and one compound, NG1, demonstrated potent activity against drug-resistant N. gonorrhoeae in both lab and mouse studies by targeting a protein critical for bacterial membrane synthesis, representing a novel mechanism of action.

Second-Round Study Uses Generative AI To Explore Novel Chemical Space

In a follow-up study, researchers leveraged generative AI to design entirely new molecules targeting the Gram-positive bacterium S. aureus. Using the CReM and F-VAE algorithms, the team allowed the AI to generate compounds without fragment constraints, guided only by the chemical rules governing atom combinations

This AI-driven approach produced over 29 million candidate molecules. The team then applied computational filters to remove compounds predicted to be toxic, unstable, or similar to existing antibiotics, reducing the pool to approximately 90 viable candidates

Of the 22 molecules that could be synthesized and tested, six displayed strong antibacterial activity against multi-drug-resistant S. aureus in laboratory assays. The leading compound, DN1, successfully cleared MRSA skin infections in a mouse model

The AI’s ability to autonomously explore vast chemical space enabled the discovery of molecules with novel mechanisms, broadly disrupting bacterial cell membranes rather than targeting a single protein

Phare Bio, a nonprofit partner in the Antibiotics-AI Project, is now optimizing NG1 and DN1 for further preclinical studies. The research team intends to apply these AI-driven design platforms to other pathogens, including Mycobacterium tuberculosis and Pseudomonas aeruginosa.

While bacterial resistance continues to outpace existing treatments, the study demonstrates that AI can explore previously uncharted areas of chemical space, offering opportunities to shift antibiotic development from reactive responses to strategic, proactive design