Generative active learning guided discovery of a tobramycin and ciprofloxacin adjuvant against multidrug-resistant Pseudomonas aeruginosa infections

Biofilm-mediated resistance in multidrug-resistant (MDR) Pseudomonas aeruginosa infections severely compromise antibiotic efficacy in clinical applications. Antibacterial adjuvants represent a promising strategy to restore antibiotic sensitivity and reduce therapeutic dosages. To identify new antibacterial adjuvants with unique structure and mechanism, we established a generative active learning workflow integrating an in-house compound repository of 725 biofilm inhibitors and a library of potential antibiofilm targets. The most potent compound STY17 was identified with sub-micromolar antibiofilm activity (IC 50 = 0.29 ± 0.01 μmol/L). In clinically isolated MDR Pseudomonas aeruginosa , STY17 significantly inhibited biofilm formation, potently synergized with tobramycin and ciprofloxacin, and suppressed the resistance development of these antibiotics. Furthermore, mechanistic studies indicated that STY17 inhibited succinate dehydrogenase to disrupt biofilm formation. In vivo , STY17 significantly enhanced the antibacterial activity of tobramycin and ciprofloxacin in Galleria mellonella and the mouse wound infection model with favorable safety profiles. These findings validated the utility of machine learning to discover novel antibacterial adjuvants, revealing STY17 as a promising candidate for antibacterial adjuvants against MDR Pseudomonas aeruginosa infections. A generative active learning workflow and further optimization resulted in a potential antibacterial adjuvant STY17, which inhibited succinate dehydrogenase and potentiated tobramycin/ciprofloxacin efficacy against multi-drug-resistant Pseudomonas aeruginosa infections.