Additional file 1 of Iterative machine learning-based chemical similarity search to identify novel chemical inhibitors

Additional file 1: Figure S1. The duplicated dose response curves to determine Kd values of chemical compounds are shown for MEK1, MEK2, and MEK5. X-axis represents ligand concentration (nM) and Y-axis relative inhibitory activity by KdELECT service. Figure S2. Structurally similar molecules were identified via substructure search in Reaxys database. Figure S3. Molecular docking conformations of ZINC5814210 for MEK1, MEK2, and MEK5 are superimposed with ATP found in the MEK1 structure (PDB ID: 3V01). Figure S4. Two-dimensional interaction diagram of previously reported MEK1 inhibitors retrieved by 2D fingerprint similarity. Figure S5. Two-dimensional interaction diagram of MEK-ZINC5479148 docking models. Figure S6. Two-dimensional interaction diagram of MEK-ZINC32911363 docking models. Except MEK2, ZINC32911363 has better Kd binding affinity to other two MEKs. Figure S7. The molecules selected based on binding free energy scores from either MM/GBSA or MM/PBSA or both of the methods. Table S1. Experimental chemical activity data and cross-validation results (AUC of Precision-Recall curve) for each test target protein. Table S2. Comparison of the prediction performance of the standard single chemical-based Random Forest model with the ECBS model trained with PP-NP-NN data. Table S3. Estimation of chemical pair data size. Table S4. LogP values for the tested compounds. Table S5. GNINA docking scores for MEKs are shown with biochemical binding affinity data in Table 3. Table S6. The target prediction results for ZINC5814210 from Swiss target prediction server. Table S7. The target prediction results for ZINC5814210 from Structure Ensemble Approach (SEA) server