Artificial intelligence-guided phenotypic drug repurposing against Streptococcus pneumoniae

The prevalence of antimicrobial resistance (AMR) within the common bacterial pathogen Streptococcus pneumoniae makes it a priority for the development of new antibiotics. While artificial intelligence (AI) has recently boosted phenotype-based repurposing of drugs for other human pathogens, this remains to be investigated for S. pneumoniae. Thus, we leveraged ensembles of transformer, graph and tree models, each trained on a set of 1,898 actives along with 39,266 inactives, to prospectively examine 6,747 drugs. Of 11 selected candidate antibiotics, nine were found to strongly reduce in vitro growth of S. pneumoniae R6, with MIC₅₀ values of ≤ 0.4 μg/mL. The most potent drugs, thiostrepton and ceftiofur, had MIC₅₀ values of 0.0001 μg/mL (60.1 pM) and 0.0004 μg/mL (764 nM), respectively. Thiostrepton remained highly potent even against multidrug-resistant strains, suggesting it could be effectively deployed to treat common non-invasive S. pneumoniae infection as part of antibiotic stewardship efforts.