Figure 1.
Binding energy landscape evaluation for docking simulation of Chk1 protein kinase inhibitor (PDB HET ID: 422) to Chk1protein kinase (PDB ID: 2br1).
Color-coded representation of ICM Docking Scores of all generated docking poses and evaluation of the energy gap. Each dot represents one docking pose of Chk1 protein kinase inhibitor. The dots are colored in different colors spread evenly among rmsd-values from 1.1 to 11.8 Å to the crystallographic position of Chk1 protein kinase inhibitor (PDB HET ID: 422). The lowest ICM Docking Score = −36.8 kcal/mol (1.1 Å (rmsd) to the crystallographic position), the highest ICM Docking Score = −5.2 kcal/mol (10 Å (rmsd) to the crystallographic position), and the energy gap of −21.5 kcal/mol.
Figure 2.
Distributions of AUC values obtained from single-receptor conformation VS.
(A) for 485 protein conformations included in the Pocketome benchmark and (B) for 485 protein conformations split in 334 holo and 151 apo structures. See Tables S1, S2, S3 for details.
Figure 3.
Distributions of AUC values obtained from multiple-receptor conformation VS.
(A) for 40 protein domains included in the Pocketome benchmark and (B) for 40 protein domains split in 40 holo and 36 apo domains. See Tables S4, S5, S6 for details.
Figure 4.
Color-coded dependence of achieved improvement in ROC AUC values obtained by the energy gap on original ROC AUC values obtained by docking score.
Each dot represents one protein domain. The dots are colored in different colors spread evenly among P-values from 0.00 to 1.00 (smallest P-values, that is, most statistically significant are red, highest P-values are blue). Overall improvement is achieved in 77% of cases in binder/decoy ligand discrimination. In 14% of cases the energy gap and the docking score performed equally. The improvement is particularly high (up to 0.19 AUC units) for the previously problematic cases with original AUC value below 0.9. In fact for these cases we obtained improvement in 83% of the proteins domains.
Table 1.
The performance of the energy gap obtained from MRC VS for 36 protein targets included in the DUD benchmark.
Figure 5.
Distribution of AUC values obtained from single-receptor conformation VS for 485 protein conformations included in the Pocketome benchmark when different energy cutoffs from the best docking score were applied to cluster binding conformations within the native binding phase.
See Table S7 for details.
Table 2.
The performance of the energy gap obtained from single-receptor conformation VS for 485 X-ray protein conformations included in the Pocketome benchmark when different energy cutoffs from the best docking score were applied to cluster binding conformations within the native binding phase.
Figure 6.
Changes in the success rate of finding the correct X-ray binding mode as a function of energy cutoff.
aAccording to ICM scoring function.
Figure 7.
Histogram of the difference in the ranking of PERK inhibitors by the energy gap and the best docking score.
“0” indicates no change in ranking by the energy gap as compared to the best docking score: for example, “0” means that ligand X was the 11th ranked compound in the list by the energy gap and also the 11th ranked compound in the list by the best docking score. Negative numbers mean that the energy gap ranking is higher than the best docking score ranking: e.g. if the compound is ranked 5th in the list by the energy gap and 7th in the list by the best docking score the above score would be −2. The histogram shows many more compounds with negative difference scores showing that the energy gap results in a higher true positive yield upon experimental testing of the top N compounds in this case of VS against protein homology models that was independent of the set in this study.