Deep learning-enabled discovery of antibiotics effective against Neisseria gonorrhoeae

ABSTRACT: Neisseria gonorrhoeae is a Gram-negative, sexually transmitted pathogen that poses a major public health threat due to rapidly increasing resistance to all recommended antibiotics. Addressing this crisis requires more efficient approaches to antibiotic discovery and the replenishment of the dwindling drug development pipeline. Here, we demonstrate that deep learning models can augment high-throughput screening to identify readily available molecules with narrow-spectrum activity against multidrug-resistant N. gonorrhoeae. We phenotypically tested 38,650 small molecules for growth inhibition and used these data to train a predictive graph neural network (GNN). Benchmarking against alternative architectures, including large language models, revealed that GNNs most effectively identified active, drug-like molecules that were structurally distinct from both the training set and known antibiotics. Applying the model to ~6 million compounds in silico, we prioritized 213 for experimental testing and found that 83 (38%) inhibited N. gonorrhoeae growth. Two compounds were structurally novel, potent against all tested multidrug-resistant strains, displayed favorable selectivity indices, and were rapidly bactericidal with low frequencies of resistance. Multi-omics analyses revealed that these compounds circumvent resistance by targeting previously unexploited pathways in N. gonorrhoeae. Our findings establish a paradigm for deep learning–enabled discovery of selective antibacterial agents and provide a promising path toward addressing the urgent threat of antimicrobial resistance in N. gonorrhoeae.

INSTRUMENT(S):

ORGANISM(S): Neisseria Gonorrhoeae (strain Atcc 700825 / Fa 1090)

SUBMITTER: Massimiliano Gaetani  

LAB HEAD: Massimiliano Gaetani

PROVIDER: PXD067696 | Pride | 2026-05-07

REPOSITORIES: Pride