Fig 1.
Overview of the ReplicaDock2 protocol.
Starting from an initial docking pose i.e. a structural model with randomly oriented protein partners, the protocol perturbs the protein partners and slides them into contact. This creates an initial docking pose for the low-resolution stage. Here, the pose object is copied to three parallel replicas per trajectory, and each replica performs rigid body moves (rotation-translation) and backbone moves for each MC trial, followed by exchange between replicas after every 1,000th trial. Each exchange obeys the Metropolis acceptance criterion and if accepted, the low resolution structure is output. Each trajectory completes 2.5 × 105 MC trial steps, and produces ∼5,000 candidate structures. Lastly, all produced structures undergo an all-atom refinement comprising of side-chain packing, small rigid-body motions, and energy minimization to output final docked structural models.
Fig 2.
T-REMC improves low-resolution performance in global rigid-body and local flexible docking for two representative protein targets.
(A) Global rigid-body docking performance for protein targets 2CFH (trafficking protein particle complex subunits) [37] and 1XQS (HspB1 core domain complexed with Hsp70 ATPase domain) [38]. Plots show the Motif Updated Dock Score (REU) vs all-atom Cα rmsd (Å). Blue points denote the refined native structures. (B) Comparison of different residue selections for performing backbone moves. Performance of ReplicaDock 2.0 with four conditions: (1) 5.5 Å interface patch, (2) 8 Å interface patch (3) 5.5 Å interface patch + loops, (4) 8 Å interface patch + loops. The metric is 〈N5〉, the average number of near-native models in the five top-scoring structures. For reference, RosettaDock 4.0 performance is highlighted in gray. (C) Local flexible backbone docking performance. Motif Updated Dock Score (REU) vs Cα rmsd (Å) for two targets, 2CFH [37] and 1XQS [38]. Panels show ∼5,000 decoys generated by RosettaDock 4.0 (left) and ReplicaDock 2.0 (right, this work).
Fig 3.
Improvement in docking performance after full protocol for two representative targets.
(A,D) Interface score (REU) vs I-rmsd (Å), (B,E) Ligand-RMSD(Å) versus Receptor-RMSD(Å), and (C,F) Interface score (REU) vs fraction of native-like contacts post all-atom refinement for RosettaDock 4.0 [11] and ReplicaDock 2.0(this work) for two targets 2CFH and 1XQS. Relative to RosettaDock 4.0, ReplicaDock 2.0 samples decoys that score better, are closer to the native, have higher native-like contacts(fnat) and better CAPRI quality. However, backbone RMSDs (B,E) have not moved closer to the native but rather diverged away from it.
Fig 4.
Comparison of performance metrics between RosettaDock 4.0 and ReplicaDock 2.0 for individual complexes in a benchmark set of 88 docking targets.
(A) Comparison of 〈N5〉 values after low-resolution and high-resolution stages (full protocol), respectively. Dashed lines highlight the region in which the two protocols differ significantly, i.e. by more than one point in their 〈N5〉 values. Different symbols correspond to each target’s difficulty category (circle: rigid; triangle: medium; diamond: flexible). Points above the solid line represent better performance in ReplicaDock 2.0, while points below the line represent better performance in RosettaDock 4.0. After the full protocol, 24 targets are modeled significantly better and 14 complexes are modeled significantly worse. (B) Probability density curves versus 〈N5〉 for all targets for ReplicaDock 2.0 (green) and RosettaDock 4.0 (purple). Low-resolution performance is indicated by lines (bright pink and bright green), and high-resolution performance is denoted by shaded area (purple and green). (C) Probability density curves versus full-protocol average N5 for rigid, medium and flexible targets respectively.
Fig 5.
Directed induced-fit improves flexible protein docking performance.
(top) (a,b,c) ReplicaDock 2.0 without directed backbone sampling of putative interfaces i.e. unbiased moves, finds medium-quality structures (colors: green = high quality, red = moderate quality, yellow = acceptable quality, gray = incorrect) (bottom) (e,f,g) ReplicaDock 2.0 with directed backbone sampling of mobile residues improves protein docking and obtains high-quality structures. (d,h) Comparing with the Ras’ unbound structure (grey) superimposed over the bound (green), the docked structure loop (blue) has moved closer to the bound state (green) for the two cases respectively. With directed sampling, it is able to capture the backbone structure to sub-angstrom accuracy.