Figure 1.
Each step in the pHDock workflow is colored based on the differences compared to RosettaDock: unmodified steps are colored in grey, and steps with minor (light orange) and major (dark orange) modifications are colored in shades of orange.
Figure 2.
Docking predictions for xylanase – TAXI-IA complex.
Docking plots generated by (A) RosettaDock, and (B) pHDock at pH 4.6. Grey, orange, and red points represent incorrect, acceptable-, and medium- quality predictions, respectively. Discrimination scores are shown in the bottom right corner of the plots. (C) Interface of the top-scoring pHDock prediction (medium accuracy) superimposed on the crystal complex (grey) (2B42 [32]). Predicted orientation of the TAXI-IA inhibitor and xylanase, cyan and green, respectively; critical His-374 residue from TAXI-IA, spheres; xylanase active site and other critical binding site residues, sticks.
Figure 3.
Summary of pHDock performance.
Correlation plot comparing discrimination scores of pHDock and RosettaDock docking predictions for each target in the complete benchmark dataset. Complexes docked at acidic pH (pH≤7.0) and basic pH (pH>7.0) are represented as circles and squares, respectively. The discrimination score cutoffs for a successful prediction (D<0) are marked using broken lines. Corner numbers indicate the total predictions in each plot section (edges defined by the broken lines and the solid line at 45°).
Figure 4.
Distribution curves of interface RMSDs (Irmsd) and fraction of recovered native contacts (fnat) for the docking models.
(A) Irmsd distribution curve of the lowest-Irmsd models generated using pHDock (orange) and RosettaDock (grey). (B, C) Irmsd and fnat distribution curve for the top-ranked models according to interface scores (Isc) for each protein complex. The distribution curves are generated after independent sorting of the pHDock and RosettaDock models based on (A, B) increasing Irmsd values and (C) decreasing fnat.
Figure 5.
Distributions of native and model interface hydrogen bonds.
Kernel density estimate curves for the number of (A) interface hydrogen bonds and (B) interface hydrogen bonds involving ionizable residues in the top-scoring models generated using pHDock (orange) and RosettaDock (grey), and the native crystal complexes (black) across the complete Docking Benchmark dataset. Frequency histograms of the fraction of (C) recovered interface hydrogen bonds and (D) recovered interface hydrogen bonds involving ionizable residues in the top-scoring models.
Figure 6.
Hydrogen bonding recovery correlates with docking performance.
Docking plots generated using RosettaDock and pHDock for (A) tumor susceptibility gene 101 protein–Ubiquitin complex (1S1Q; pH 4.6), (B) PPARgamma+RXRalpha–GW409544+co-activator peptide complex (1K74; pH 7.5), and (C) CDK2 kinase–cell cycle-regulatory protein CksHs1 complex (1BUH; pH 7.5). Grey, orange, red, and blue points represent incorrect, acceptable-, medium-, and high-quality models, respectively. Discrimination scores are shown in the bottom right corner of the plots. The right panel shows structures of the top pHDock (blue) and RosettaDock (green) models superimposed on the native complex (red). The number of native hydrogen bonds among the total interface hydrogen bonds observed in the bound crystal complex, and the top-scoring pHDock
and RosettaDock
models are also listed.
Figure 7.
pH-dependent binding effects in Fc–FcRn complex.
(A) Interface score of the top pHDock prediction for the Fc–FcRn complex as a function of the docking pH. (B) Interface score vs Irmsd plots generated using pHDock at pH 6.25 and pH 7.50. (C) Top pHDock models at pH 6.25 (cyan) and pH 7.50 (green) showing the three critical ionic interactions responsible for the large pH-dependent binding affinity change. Note the change in the protonation states of His-435 and His-436.