A NOVEL AI framework has successfully designed new antibiotic candidates with activity against drug-resistant Neisseria gonorrhoeae and Staphylococcus aureus, offering a promising route to address the escalating antimicrobial resistance crisis.
The rise of antimicrobial resistance has been fuelled by a lack of structurally novel antibiotics, with most discovery efforts focused on screening existing chemical libraries. In this study, researchers developed a generative AI platform combining genetic algorithms and variational autoencoders to create completely new antibacterial molecules. Two complementary approaches were used: a fragment-based strategy, which screened more than 10⁷ chemical fragments in silico against N. gonorrhoeae or S. aureus before expanding promising hits, and an unconstrained de novo design pipeline to explore uncharted regions of chemical space. Candidate compounds from both approaches were synthesised and tested for selective antibacterial activity.
Of 24 synthesised compounds, seven showed selective activity against target pathogens. Two compounds emerged as leads: one active against multidrug-resistant N. gonorrhoeae and the other against methicillin-resistant S. aureus (MRSA). Each displayed a distinct mode of action, confirmed through mechanistic assays, and demonstrated in vivo efficacy. In mouse models, the N. gonorrhoeae lead reduced bacterial burden in a vaginal infection model, while the MRSA lead was effective in a skin infection model. Structural analogues of both leads also retained antibacterial properties, supporting the robustness of the design approach.
These findings demonstrate that generative deep-learning methods can go beyond repurposing known molecules, instead creating novel antibiotic structures with unique mechanisms of action. The study’s platform opens the door to systematic exploration of unexplored chemical space, potentially accelerating the discovery of urgently needed antibiotics.
Reference
Krishnan A et al. A generative deep learning approach to de novo antibiotic design. Cell. 2025; DOI:10.1016/j.cell.2025.07.033.
