3.3. Stepwise Structure-Based Virtual Screenings

Structure-based virtual screenings were performed using rDock (2013) [27] and AutoDock Vina version 1.1.2 [28]. Both interfaces are available in the Docking Study with HyperChem (DSHC) software I developed [20,29], and the resulting docking modes filtered by the rDock score threshold were more precisely simulated using AutoDock Vina.

Prior to the docking simulations, favipiravir-RTP was removed from the RdRp complex:dsRNA system. Then, docking simulations for the 3D structure database (1,838,257 compounds) were performed using the relatively reliable, high-speed docking simulation program rDock under the default condition. The cavity condition for the rDock docking simulations was prepared using favipiravir-RTP (the reference ligand method) under a default condition. The docking simulations outputted three docking modes per trial compound in the SDF file format. Consequently, 5,507,523 docking modes were obtained (rDock could not assess or support 2416 out of 1,838,257 compounds). These docking modes were filtered by the rDock score threshold of ≤−60 kcal·mol⁻¹ to obtain 88,250 distinct compounds (168,954 docking modes) in the SDF file format. The ChEMBL IDs of these distinct compounds were subjected to KNIME version 4.1.2 [30] to collect the compound information from the ChEMBL web server. Some information was manually collected from the Kyoto Encyclopedia of Genes and Genomes database [31].

From the 168,954 docking modes obtained via the virtual screenings, the 88,250 distinct hit compounds had two docking modes on average. The hit compounds, including the 249 drugs I found, were more precisely investigated using the AutoDock Vina docking simulations with these docking modes as the initial structures. Subsequently, the resulting 168,954 docking modes were separated and converted into individual PDBQT files using DSHC (some compounds, including unsupported metal atoms, such as boron and selenium, were rejected at this stage). Then, more precise docking simulations were performed using the AutoDock Vina In Silico Screenings Interface software integrated into DSHC. The aforementioned RdRp complex:dsRNA system in the PDB file format was also converted to a PDBQT file using DSHC. A configuration file with cavity information (the value of the grid box was set to center_x = 100.646 Å, center_y = 96.712 Å, center_z = 112.158 Å with size_x = 22.254 Å, size_y = 24.537 Å, and size_z = 22.606 Å) was prepared using DSHC, and other docking conditions were set to the default values (the top nine docking modes per trial compound were maximally outputted). Docking simulations with AutoDock Vina produced 1,508,525 docking modes, which were filtered by the AutoDock Vina score (empirical binding free energy) threshold of −12 kcal·mol⁻¹. Because the AutoDock Vina score reflects empirical binding free energy, I expected that a score of −11 kcal·mol⁻¹ would theoretically represent a subnanomolar order of binding affinity with the RdRp complex:dsRNA. When the threshold for screening was set to less than this value, I obtained 2259 distinct compounds (6560 docking modes) as hit compounds. To more realistically concentrate the number of hit compounds, I set the threshold value to ≤−12 kcal·mol⁻¹. Consequently, I obtained 184 distinct compounds (353 total docking modes). The ChEMBL IDs of these distinct compounds were subjected to KNIME to collect the compound information from the ChEMBL web server.