AutoDock-SS: AutoDock for Multiconformational Ligand-based Virtual Screening

Ligand-based virtual screening (LBVS) is widely employed in drug discovery to identify potential leads when the crystallographic structure of the target protein remains unknown. In this study, we introduce a novel three-dimensional LBVS workflow incorporating newly designed ligand-based grid maps and AutoDock-GPU, referred to as AutoDock-SS (Similarity Searching). AutoDock-SS features two modes supporting either single query ligands or multiple pre-aligned ligands as input. The virtual screening performance of AutoDock-SS single-reference mode was evaluated using the standard Directory of Useful Decoys – Enhanced (DUD-E) dataset, with the method outperforming alternative state-of-the-art 3D LBVS methods. The mean area under the receiver operating characteristics curve (AUC) reached 0.775 (exceeding the maximum of 0.755 for alternative methods), and the enrichment factor at one percentage (EF1%) was 25.72. AutoDock-SS multi-reference mode was assessed on the augmented DUD-E dataset (DUD-E+) using five pre-aligned query ligands, displaying statistically significant superior prediction accuracy (mean AUC of 0.843, p < 10-5) and higher mean EF1% (34.59) compared to the singe-reference mode. Overall, the findings suggested that AutoDock-SS exhibits enhanced binding conformation prediction when screening the original DUD-E dataset. The proposed method accounts for the shape, pharmacophore, and electrostatic potential of the query ligand(s) to improve prediction accuracy and includes a built-in conformational sampling algorithm that natively considers compounds as naturally conformationally flexible—an aspect not present in existing LBVS methods.