Replica exchange improves sampling in low-resolution docking stage of RosettaDock

Abstract

Many protein-protein docking protocols are based on a shotgun approach, in which thousands of independent random-start trajectories minimize the rigid-body degrees of freedom. Another strategy is enumerative sampling as used in ZDOCK. Here, we introduce an alternative strategy, ReplicaDock, using a small number of long trajectories of temperature replica exchange. We compare replica exchange sampling as low-resolution stage of RosettaDock with RosettaDock's original shotgun sampling as well as with ZDOCK. A benchmark of 30 complexes starting from structures of the unbound binding partners shows improved performance for ReplicaDock and ZDOCK when compared to shotgun sampling at equal or less computational expense. ReplicaDock and ZDOCK consistently reach lower energies and generate significantly more near-native conformations than shotgun sampling. Accordingly, they both improve typical metrics of prediction quality of complex structures after refinement. Additionally, the refined ReplicaDock ensembles reach significantly lower interface energies and many previously hidden features of the docking energy landscape become visible when ReplicaDock is applied.

Figure 1. Detailed analysis of individual docking stages on bound target 1sq2.

A) Interface RMSD (I_rms) before and after the Monte-Carlo optimization in the low-resolution stage of RosettaDock's shotgun sampling, B–C) I_rms before and after all-atom refinement for shotgun and ReplicaDock sampled ensembles, respectively. The colorbar indicates the density of data points at given position of the scatter plot, A–C) use a same colorbar range. The insets show the distribution of differences between I_rms after and before the respective sampling stage has been applied (negative values reflect an improvement in I_rms).

Figure 2. Detailed analysis of shotgun and ReplicaDock sampling on target 1ppf.

A) energy distribution of shotgun sampling generated low-resolution decoys. B) energy distribution of conformations sampled by ReplicaDock at respective inverse temperatures. C–F) Population of sampled conformations in spherical coordinates. Partner A is fixed at the center and the position of Partner B with respect to an idealized spherical surface around Partner A is recorded. The native structure is labeled as white dot (arrow in C). G–J) Conformations are assigned to grid-cells as in C–F, but shown is the lowest energy of all conformations assigned to the respective grid cell. The same color-scale is used for each plot of a row, and the colorbars are attached to the rightmost panel.

Figure 3. Fraction of hits in low-resolution docking.

Conformation with I_rms ≤2.5 Å to the bound complex are considered as a hit (Section 3.9). Blue, green and red represent the results of shotgun sampling, ZDOCK and ReplicaDock, respectively.

Figure 4. Interface RMSD vs. Interface Energy after refinement on target 1ppf and 1mlc.

A) and C) refinement of shotgun sampling generated ensembles, B) and D) refinement of ReplicaDock generated ensembles. The red dots represent the RelaxedNative ensembles (Results).

Figure 5. Interface RMSD vs. Interface Energy after refinement for all the 30 unbound docking targets.

The red dots represent the RelaxedNative ensembles(Results). The interface RMSD is shown on the x-axis, the interface energy on the y-axis. The same energy range is used for displaying both, shotgun sampling (blue) and ReplicaDock (black), results of each target, respectively. The vertical gray lines correspond to I_rms of 5.0 Å, and the two horizontal gray lines correspond to interface energy −4 and −8 Rosetta Energy Units.

Figure 6. Interface energy distribution obtained with different simulation protocols.

The energies for each target and method have been normallized to the dynamic range of interface energies observed for the respective target across all methods. The interface energies are normalized by the absolute value of mean energy of the 10 lowest observed energies for this target (highest energy is always 0). The x-percentile energy is the scaled energy value that separates off x% of the lowest energy decoys for a given simulation result. Shown are the distributions of x-percentile energies across all 30 targets for a) the 5%-tile, b) the 1%-tile and c) the 0.1%-tile, respectively.

Figure 7. The centroid energy function prefers an alternative binding modes for the bound docking target 1emv.

A–B) shotgun and ReplicaDock sample the low-resolution docking stage with ‘capped’ centroid energy function. C–D) shotgun and ReplicaDock sample the low-resolution docking stage with centroid energy function with no cap. The structure indicated by the red circle will be shown in Figure 8.

Figure 8. Electrostatic surface potential of native (A–D) and interchain_cen preferred conformation (E–H) in all-atom representation.

A) native complex with receptor in green and ligand in red B) electrostatics map of the native complex C) electrostatics map of native ligand interface D) electrostatics map of native receptor interface E) interchain_cen preferred conformation after refinement F) electrostatics map G) electrostatics map of receptor interface H) electrostatics map of ligand interface. The yellow lines in C, D, G and H indicate the respective interface regions, and number pairs (e.g. 1 in C and 1′ in D) indicate corresponding contact regions. Relations of viewing angle are given between panels where required.