Can AI Help Accelerate the Development of Antibiotics for Gonorrhea and MRSA?

Promising Fragments Target Gonorrhea

The study tested two different approaches to designing new antibiotics using generative AI trained on a library of known compounds that inhibit the growth of various bacterial species. The first approach was based on screening a library of approximately 45 million chemical fragments composed of all possible combinations of 11 atoms of carbon, nitrogen, oxygen, fluorine, chlorine, and sulfur, together with additional known fragments from the Enamine REAL library. The AI selected fragments showing antimicrobial activity and used them as the basis for designing new molecules.

The researchers used machine-learning models trained to predict antimicrobial activity against the causative agent of gonorrhea, the bacterium Neisseria gonorrhoeae. A set of approximately 4 million candidate fragments was then narrowed down by removing those that could be cytotoxic to human cells or were similar to existing antibiotics.

This left roughly one million candidates. Through several rounds of further experiments and computational analysis, the researchers identified a fragment they named F1, which appeared to be effective against N. gonorrhoeae. This fragment was then used as the basis for designing additional compounds using two different artificial intelligence algorithms. These generated approximately 7 million candidates containing F1, which were further tested in silico. A final set of 80 of the most promising compounds was passed to organic chemists, who were able to synthesize only two of them.

The compound NG1 ultimately proved to be highly effective against N. gonorrhoeae in both in vitro experiments and a mouse model of resistant gonococcal infection. Further experiments revealed that NG1 interacts with the protein LptA, which is involved in the synthesis of the bacterial outer membrane. NG1 appears to disrupt membrane synthesis, which is lethal to bacterial cells.

Targeting MRSA

The researchers then tested the potential of generative artificial intelligence to freely design molecules targeting the elimination of the Gram-positive bacterium Staphylococcus aureus. In this case, the only constraints were general chemical rules governing how atoms form molecules. The AI generated more than 29 million compounds, from which 90 candidates with potential activity against S. aureus were selected using the same principles applied in the screening against N. gonorrhoeae.

Twenty-two of these compounds were successfully synthesized, with six showing strong antibacterial activity against multidrug-resistant S. aureus. The most successful compound, DN1, was able to cure a skin infection caused by methicillin-resistant S. aureus (MRSA) in a mouse model. These molecules also appear to disrupt bacterial cell membranes, but with broader effects that are not limited to interaction with a single specific protein.

A Weapon in the Fight Against Superbugs

According to J. Collins, Professor of Medical Engineering at MIT, the study demonstrated that artificial intelligence can enable researchers to propose new molecular designs quickly and at low cost, giving them a much-needed advantage in the battle against superbugs.

The researchers are now further investigating analogs of NG1 and DN1 and are also working on modifications that would allow for additional preclinical testing. They plan to continue using AI platforms to search for treatments against other pathogens of interest, particularly Mycobacterium tuberculosis and Pseudomonas aeruginosa.

Editorial Team, Medscope.pro

Sources:
1. Krishnan A., Anahtar M. N., Valeri J. A. et al. A generative deep learning approach to de novo antibiotic design. Cell 2025 Aug 7: S0092-8674(25)00855-4, doi: 10.1016/j.cell.2025.07.033.
2. Trafton A. Using generative AI, researchers design compounds that can kill drug-resistant bacteria. MIT News, Aug 14, 2025. Available at: https://news.mit.edu/2025/using-generative-ai-researchers-design-compounds-kill-drug-resistant-bacteria-0814