Artificial intelligence-guided discovery of a lead compound with antifolate-like activity against bacterial and human thymidylate synthases.

Thymidylate synthase (TYMS) is an established colorectal cancer (CRC) target. We evaluated whether conserved active-site architecture between human TYMS and the bacterial homolog TYSY KLEP7 could support early cross-species TYMS-axis lead discovery within the antifolate landscape. AlphaFold3, Chai-1 complex modeling, crystal-structure cross-validation, docking, ligand strain energy, MM-GBSA rescoring, and explicit-solvent MD were integrated to compare pocket states and prioritize antifolate-like Compound 8. Modeling indicated that Compound 8 preserves an acidic folate-groove anchoring motif while tolerating species- and ligand-induced pocket variation. Phenotypic assays showed dose-dependent inhibition of CRC cell viability with preliminary NCM460 selectivity, whereas antibacterial activity against K. pneumoniae was modest, with a MIC of 96 µM and bactericidal activity only at 4× MIC after 24 h. Target-axis assays supported TYMS/thyA engagement: siTYMS and AS-thyA reduced transcript levels, and Compound 8 inhibited crude-lysate TS activity with apparent AC50 values of 7.8 µM in HCT116 and 13.2 µM in K. pneumoniae. Overall, this work establishes an AI-enabled structure-to-phenotype framework and positions Compound 8 as an early antifolate-like TYMS-axis lead requiring purified-enzyme, ADME, safety, and antibacterial optimization studies.