AI-Guided Phenotypic Drug Repurposing Against Streptococcus pneumoniae
Synopsis
This study applied AI-guided phenotypic drug repurposing to drug-resistant Streptococcus pneumoniae, using ensembles of transformer, graph, and tree models trained on 1849 actives and 34 503 inactives to prospectively examine 6747 drugs, selecting 11 candidate antibiotics of which nine strongly reduced in vitro growth of S. pneumoniae R6 (IC50 ≤ 0.4 µg/mL), with the most potent drugs thiostrepton and ceftiofur showing IC50 values of 0.0001 µg/mL (60.1 pM) and 0.0004 µg/mL (764 pM), respectively, and thiostrepton remaining highly potent against multidrug-resistant strains.
Overview of the AI‐driven drug repurposing workflow against Streptococcus pneumoniae cultures. (a) The dataset was created from bioassays from PubChem and ChEMBL, filtered specifically by keywords related to pneumococci and resistance, and further refined to include only bioassays with specific assay types such as MIC, MIC 90 , MIC 50 , and Activity. (b) This study used three AI models to prospectively evaluate 6747 small‐molecule drugs for their potential to inhibit S. pneumoniae , selecting those with high consensus predictions as promising repurposing candidates. (c) MoLFormer models are adapted for antibiotic screening by adding a trainable Feed‐Forward Network (FFN) block (highlighted in orange). (d) From the top predictions, we selected 11 compounds for in vitro testing based on high model scores and chemical structure dissimilarity to training molecules.
PubMedInterpretation
Applied AI-driven phenotypic drug repurposing to S. pneumoniae, a pathogen for which this strategy had not yet been investigated. The text states that AI has boosted phenotype-based repurposing for other human pathogens but that this 'remains to be investigated for S. pneumoniae'; this study addresses that gap. Method is explicit: ensembles of transformer, graph, and tree models trained on 1849 actives and 34 503 inactives, prospectively examining 6747 drugs.
Selected 11 candidate antibiotics from 6747 drugs, of which nine strongly reduced in vitro growth of S. pneumoniae R6. Provides a concrete candidate list for S. pneumoniae rather than stopping at computational prediction. In vitro validation: 9 of 11 candidates showed strong growth reduction with IC50 ≤ 0.4 µg/mL.
Identified two highly potent drugs: thiostrepton (IC50 0.0001 µg/mL, 60.1 pM) and ceftiofur (IC50 0.0004 µg/mL, 764 pM). Reports picomolar-level potency data, offering high-priority starting points for further development. In vitro IC50 measurements with explicit, very high potency values.
Thiostrepton remained highly potent against multidrug-resistant strains, suggesting it could be deployed to treat common non-invasive S. pneumoniae infection as part of antibiotic stewardship. Extends the candidate's value from susceptible strains to multidrug-resistant strains and links it to antibiotic stewardship practice. The text states thiostrepton 'remained highly potent even against multidrug-resistant strains'; this is an in vitro potency observation, and the authors frame the application prospect with 'suggesting'.
Perspective
The study targets the specific pathogen S. pneumoniae and uses in vitro phenotypic screening with IC50 measurements, suited to discovering and prioritizing candidate antibiotics; its most direct audience is anti-infective drug discovery and antibiotic stewardship research. The text suggests thiostrepton could be deployed to treat common non-invasive S. pneumoniae infection, but this is the authors' proposal based on in vitro results.
The loaded text is abstract-level and does not provide model performance metrics, candidate selection thresholds, experimental replicates, statistical methods, or in vivo validation data; the potency of thiostrepton against multidrug-resistant strains is described only in words, without specific strains or values. Readers may watch for these details in the full paper and for evidence translating in vitro potency into in vivo efficacy.