Multi-target in silico screening of African natural products (AfroDb) against methicillin-resistant Staphylococcus aureus: a benchmark-calibrated assessment

This archive contains the complete computational record of a multi-target virtual screen of AfroDb, a library of natural products from African medicinal plants, against four mechanistically distinct Staphylococcus aureus drug targets: penicillin-binding protein 2a (PDB 1MWU), dihydrofolate reductase (2W9H), enoyl-ACP reductase FabI (3GR6) and the ATPase domain of DNA gyrase subunit B (3TTZ). The pipeline ran in four stages deduplication and Lipinski plus lead-likeness filtering of 903 unique structures down to 155 lead-like candidates; AutoDock Vina docking of those candidates alongside one clinically used reference inhibitor per target (156 ligands per target); ADMET prediction with ADMET-AI; and flow-limited physiologically based pharmacokinetic simulation of tissue exposure for the resulting shortlist. The archive provides the input library and prepared receptor files, a single machine-readable parameters.json recording every threshold and docking box, the analysis scripts, the raw docking-engine output for every run, the ADMET and PBPK result tables, the publication figures, a pinned software environment and an md5 file manifest. Reference inhibitors were embedded in each ranking as an explicit calibration benchmark: calibration succeeded for FabI (triclosan, rank 2 of 156) and failed for DHFR (trimethoprim, rank 141 of 156), so the four target columns in this dataset do not carry equal evidential weight and should not be averaged