AI-Driven Discovery of Prototype CLEC4M Inhibitors Targeting Marburg Virus Entry via Integrated Machine Learning and...

Marburg virus (MARV), a highly pathogenic member of the Filoviridae family, causes severe hemorrhagic fever with a high case fatality rate and currently lacks effective therapeutics. The viral entry process, mediated by the interaction between the MARV glycoprotein (GP) and host receptor C-type lectin domain family 4 member M (CLEC4M) (L-SIGN), represents a critical target for early-stage intervention. The active compounds from BindingDB and the decoy from DUDE were used. The RDKit was used for feature engineering. Machine learning models were trained on an initial dataset consisting of 56 active chemicals and 1232 decoys. Among the tested algorithms, the Random Forest model demonstrated superior performance, achieving the highest discriminative ability (AUC = 0.93, MCC = 0.88) on the test set. Virtual screening of 11,032 phytochemicals resulted in 120 predicted actives, of which 42 compounds satisfied drug-likeness criteria. Subsequent molecular docking identified three lead compounds (PubChem IDs: 42608095, 5281601, and 11243993) with moderate-to-promising binding affinities (-6.3 to -6.5 kcal/mol) toward the CLEC4M binding site. ADMET analysis revealed favorable pharmacokinetic and toxicity profiles for the selected lead compounds. DFT calculations of the three compounds highlighted their electronic stability and reactive nature, indicating that PubChem IDs 42608095 and 5281601 possess particularly stable electronic properties conducive to favorable target interactions. Combining machine learning models with molecular docking and Molecular Dynamics (MD) simulations worked well in finding promising phytochemical inhibitors. The MM/GBSA binding free energy calculations further confirmed binding affinities, with values of -10.83 and -11.08 kcal/mol, respectively, suggesting favorable complex stability. These findings provide a pathway for developing new antiviral agents against MARV, pending further experimental validation and optimization.