Machine learning-guided QSAR screening of fluconazole analogs and FDA-approved drugs against Candida albicans, with docking, molecular dynamics, and ADMET analysis.

Invasive candidiasis caused by Candida albicans is a critical-priority fungal disease associated with high mortality, and the increasing resistance to fluconazole underscores the urgent need for new antifungal agents. In this study, a machine learning (ML)-guided quantitative structure-activity relationship (QSAR) workflow was developed to prioritize fluconazole analogs and FDA-approved drugs with predicted antifungal activity against Candida, followed by a structure-based exploration of a cell wall target. Bioactivity (IC50) data for the CHEMBL366 target were curated to 1000 compounds, encoded using Morgan fingerprints (ECFP4, 2048 bits), and used to benchmark 42 regression algorithms; the three best-performing tree ensemble models (Extra Trees, Gradient Boosting, and Bagging) were optimized with Optuna. The Extra Trees model achieved the best performance (test R2 = 0.70; RMSE = 0.38), and the applicability domain (AD) covered approximately 93% of the test compounds. Consensus virtual screening of fluconazole analogs from the ZINC database and 1683 FDA-approved drugs prioritized two structurally related analogs (ZINC000299869729 and ZINC000299869730). Molecular docking against exo-β-(1,3)-glucanase (PDB ID: 1EQP), used as an exploratory mechanistic hypothesis, produced more favorable scores (- 9.1 to - 9.6 kcal/mol) than fluconazole (- 8.5 to - 8.9 kcal/mol). In 200 ns molecular dynamics simulations, ZINC000299869730 showed greater conformational stability, while ZINC000299869729 exhibited the most favorable MM/GBSA binding free energy (- 19.02 kcal/mol). pkCSM-based ADMET profiling predicted good intestinal absorption but also shared AMES mutagenicity and hepatotoxicity alerts, which would require mitigation through structural optimization. All findings are strictly in silico and require experimental validation of antifungal activity and safety.