Additional file 1 of De novo design and bioactivity prediction of SARS-CoV-2 main protease inhibitors using recurrent neural network-based transfer learning

Additional file 1: Table S1. Validity, uniqueness and novelty (mean ± std) of SMILES generated after training. We sampled 10,000 SMILES for eachtemperature (2,000 SMILES in five independent runs). Figure S1. UMAP plot of the chemical space of scaffolds generated by the general chemical modeland scaffolds from ChEMBL (2,000 molecules were randomly selected for each set). Figure S2. Redocking experiment to validate the molecular dockingprotocol. The docked pose of ligand X77 from SARS-COV-2 Mpro (PDB: 6W79) is shown as purple sticks and the experimental binding pose as greensticks. The enzyme surface is shown in bege. The RMSD between the docked and experimental pose was 1.106 Å. Figure S3. Docked poses of LaBECFar-1and LaBECFar-3 on SARS-COV-2 Mpro. (PDB: 4MDS). The amino acid residues are shown as bege sticks and the ligands are shown as pink sticks.Figure S4. Docked poses of LaBECFar-6, LaBECFar-7 and LaBECFar-9 on SARS-COV-2 Mpro. (PDB: 6W79). The amido acid residues are shown asbege sticks and the ligands are shown as orange sticks. Table S2. FDA approved drugs predicted to be active on SARS-CoV-2 Mpro.