Figure 1.
Orientational Matching Diagram.
A toy example illustrating the matching sphere orientational matching algorithm. A) Toy receptor with 4 matching spheres shown as circles and a toy ligand with 3 spheres shown as stars. B) The distance matrices constructed from these spheres are show in the upper right. C) The 2 possible orientational matches of the ligand spheres (as stars) onto the receptor spheres with a distance tolerance of 0.1 (assuming 3 matching nodes are used, in 3D this is usually 4). D) The additional two orientations produced when the distance tolerance is raised to 0.2.
Figure 2.
DOCK score effects with varying degrees of orientational sampling.
The effect of changing the desired number of match goals, or orientational samples, on the DOCK score of both the ligands (in blue) and decoys (in red). 5 comparisons of the 4 levels of match goals are shown for Glutamate Receptor Ionotropic Kainate 1 (GRIK1) in A through E.
Figure 3.
DOCK score effects with varying degrees of orientational sampling.
A) The crystal ligand from PDB Code 1VSO. The critical contacts are defined as 3 atoms from the ligand crystal structure making key polar contacts with the protein, highlighted with spheres. 4 poses of ZINC00013260 are shown in B through E, with increasing sampling going from left to right, better DOCK scores and lower critical contact RMSD (with the exception of the critical contact RMSD rising from Match Goal of 50 to 500). Protein is shown in gray, crystal ligand shown in purple and representative docked pose shown in green, with hydrogen bonds drawn according to UCSF Chimera defaults. An additional molecule, ZINC00374553, is similarly shown in subfigures F through I, with a similar trend of increasing DOCK energies and decreasing critical contact RMSD.
Figure 4.
Enrichment changes with varying degrees of orientational sampling.
A) The histogram of changes between match goals of 50 and 20000 over all 102 DUD-E systems is shown. B) At right, the histogram of which of the five match goal levels produced the best enrichment for each of the 102 DUD-E targets. For each enrichment produced by another match gal, the histogram of the differences is shown to the left.
Figure 5.
Enrichment changes with varying degrees of orientational sampling.
The effect of changing the desired number of match goals, or orientational samples, on the overall enrichment of ligands over matched decoys is shown. Three possibilities are shown, A & C) MAP Kinase-Activated Protein Kinase 2 (MAPK2), where the logAUC goes down with increased orientational sampling, above that is the difference in DOCK energies for ligands and decoys for that target. B & D) Protein Farnesyltransferase/Geranylgeranyltransferase Type I Alpha Subunit (FNTA), where the logAUC does not change significantly with increased orientational sampling, again the difference between the ligand and decoy energies is shown above in the upper right. E), GRIK1 is shown, where the logAUC goes up with increased orientational sampling.
Table 1.
Compiler Optimizations.
Table 2.
Speed and Memory Comparison between DOCK3.5.54 and DOCK3.7.
Figure 6.
Speed versus different measures for five levels of orientational sampling.
Speed measured in mean time in hours across all 102 DUD-E Targets against three measures of docking performance: Adjusted logAUC, AUC and EF1. Data shown for the full MMFF94S energy function used in ligand bulding (Green Squares) as well as the energy function with electrostatics turned off (Orange Diamonds).
Figure 7.
At left, several conformations of a ligand built with electrostatics off. At right the same ligand built with electrostatics on. The MMFF94S energies from OMEGA are shown below each pose. The bottom conformation on either side is the lowest energy conformation according to either energy function. The scales at either side are the differences in energy score from the best conformation to the shown conformation, this is the energy window used in construction.
Figure 8.
Enrichment changes with electrostatics on or off during ligand building.
A & C), the difference in DOCK scores and logROC curves for Adenosine A2A Receptor (AA2AR), B & D) Fatty Acid Binding Protein Adipocyte (FABP4). E) A histogram illustrating the changes over the entire 102 DUD-E systems (in black). Over 60 systems do better with electrostatics on, but the mean difference is 0.0 due to the more extreme differences when electrostatics off ligand builds perform better. In green bars are the changes when the energy window is increased to 30 kcal/mol for the 17 poorest performing systems. F) Gene names for the most extreme cases where electrostatics off ligand builds perform better, with their mean logAUC difference. Blue systems are proteases, 6 of the 15 DUD-E protease systems are in this table.
Figure 9.
Hydroxyl dihedral distribution.
A) Distribution of dihedral angles in radians for hydroxyls adjacent to any aromatic six-membered rings in the Cambridge Structural Database[133]. Inset images show a phenol with the dihedral angle marked, B) is an example of off-planar aromatic hydroxyls produced by DOCK3.5.54, DOCK3.6 or in this paper as the “No Reset” option, C) is the Reset Hydroxyl version. The bin from 3.1 to −3.1 is shown only at left.
Figure 10.
Effects of resetting aromatic hydroxyls.
Changes in DOCK score and logAUC across several different orientational samplings shown for 2 systems: A & C) COMT B & D) ESR2). E) Mean changes shown across the 29 systems tested with any of the 5 match goals (20000, 5000, 2000, 500, 50). The mean over the 29 DUD-E targets with hydroxyls is also shown, with the highest or lowest difference over any of the 5 match goals shown with black error bars. The worst mean difference in favor of No Reset being better was 0.24 logAUC for BRAF, the worst difference for any match goal is SAHH of 0.53 logAUC for a Match Goal of 500.
Figure 11.
Effects of changing the bump limit on docking performance and time.
A & B) Differences for varying the bump limit at a match goal of 500 are shown here for one DUD-E target, Thyroid hormone receptor beta-1 (THB). This is one of the few cases where ligands and even some decoys find scores with a higher bump limit than they do with a lower bump limit in kcal/mol. C) The timings for this run and the mean over all 102 DUD-E targets is shown. The bump limit itself does not have a large effect on the time, but using a high bump limit instead of none roughly doubles the speed of docking.