Fig 1.
Docking ability for 72 scoring functions generated by small perturbations of the Vina scoring function. The docking ability is measured as a percentage of the 122 complexes which the scoring function managed to dock correctly, compared to: (A) Pearson correlation coefficient between calculated and experimental binding affinities. (B) Average RMSD. (B). In blue, a linear regression of the data is shown, as a trend, and the Pearson correlation coefficient of the regression is shown as R.
Table 1.
Weights applied to each term in Vina and Vinardo.
Comparison of the weights applied to each energetic term (see Eq 3).
Table 2.
Values for parameters used in energetic terms.
Comparison of the parameters used in Eqs 4 to 8, values in Å.
Table 3.
Atomic radii used in Vina and Vinardo scoring functions.
CA are aromatic carbons. Values are in Å.
Table 4.
Average contribution of each interaction to the final predicted binding energy.
Determined for the 195 protein-ligand complexes found in the PDBBIND Core 2013 database. Values are expressed as percentages of final average binding energy.
Fig 2.
Docking performance for 4 different datasets.
Docking performance of Dk_scoring, Vina and Vinardo was assessed for all protein-ligand complexes in the Astex-diverse, Iridium-HT, Iridium-MT, and CSAR 2012 datasets. Results are displayed as percentage of total compounds in the dataset for which at least one pose was correctly docked (RMSD equal or lower than 2 Å with respect to the crystallized ligand pose) and ranked by the scoring function as a Top 1, Top 3 or Top 5 pose.
Fig 3.
The long range steric term in Vina causes some ligands to be incorrectly docked.
Ligands are depicted as sticks, with hydrogens colored in white, oxygen in red, nitrogen in blue, and carbons colored according to the origin of the ligand pose. The crystallized ligand pose is shown in cyan, the Vinardo top ranked pose in green, and the Vina top ranked pose in yellow. Scale bar corresponds to a distance of 2 Å. Protein surface is colored according to amino acid hydrophobicity, using the Kyte-Doolittle scale with colors ranging from dodger blue for the most hydrophilic, white for neutral residues, and orange red for the most hydrophobic residues.
Table 5.
Scoring power for each scoring function, as defined in CASF 2013.
Correlation coefficients between experimental and calculated binding affinity. In column labeled “R”, correlation coefficients are calculated by scoring crystal structures. In column labeled “Minimized-R”, energy minimization is performed before the calculation of correlation coefficients. The average RMSD of minimized structures with respect to crystallized ligands are shown in column “RMSD” (values are in Å). Results for GlideScore-SP, ChemPLP@GOLD and X-Score are taken from [11].
Fig 4.
Scatter plots comparing predicted binding affinities to experimental binding affinities.
Predicted binding affinities were calculated using each scoring function on the crystallized poses for the PDBBIND Core 2013 database. In red, the result of a linear regression is shown.
Table 6.
Ranking power for each scoring function, as defined in CASF 2013.
The 195 structures in PDBBIND Core 2013 database consist of 65 proteins with three ligands each. Ranking power is measured as “High success” when all three ligands are correctly ranked, or “Low success”, when only the top ligand is correctly ranked. Values are reported as percentages of the total 65 proteins. Results for GlideScore-SP, ChemPLP@GOLD and X-Score are taken from [11].
Table 7.
Docking power for each scoring function, as defined in CASF 2013.
Each scoring function was used to measure the energy of many different poses for each protein-ligand complex. A scoring function was considered successful when the top scoring pose was equal to or lower than 2 Å from the crystallized pose. Values are reported as percentages of the total 195 protein-ligand complexes analyzed. Results for GlideScore-SP, ChemPLP@GOLD and X-Score are taken from [11].
Table 8.
Screening power for each scoring function, as defined in CASF 2013.
Values are reported as percentages of the total 65 proteins for which each scoring function was successful (the known true binder of said protein was ranked in the top 1%, top 3% or top 5% of the 195 compounds tested). Results for GlideScore-SP, ChemPLP@GOLD and X-Score are taken from [11].
Fig 5.
Overall and early enrichment of active compounds from the DUD dataset.
28 proteins from the DUD dataset were used to dock ligands and decoys. A thick black line represents the median value, the box limits represent the first and third quartile, while whiskers represent the 9th and 91st percentile. (A) AUC is used as a measure of overall enrichment. (B) BEDROC is used as a measure of early enrichment. For protein comt, the cofactor molecule SAM was removed from the original DUD pdb file, since many inhibitors bind to the cofactor binding site.