Protein-protein docking with large-scale backbone flexibility using coarse-grained Monte-Carlo simulations

Abstract

Structure prediction of protein-protein complexes is one of the most critical challenges in computational structural biology. It is often difficult to predict the complex structure, even for relatively rigid proteins. Modeling significant structural flexibility in protein docking remains an unsolved problem. This work demonstrates a protein-protein docking protocol with enhanced sampling that accounts for large-scale backbone flexibility. The docking protocol starts from unbound x-ray structures and is not using any binding site information. In docking, one protein partner undergoes multiple fold rearrangements, rotations, and translations during docking simulations, while the other protein exhibits small backbone fluctuations. Including significant backbone flexibility during the search for the binding site has been made possible using the CABS coarse-grained protein model and Replica Exchange Monte Carlo dynamics. In our simulations, we obtained acceptable quality models for the set of 12 protein-protein complexes, while for selected cases, models were close to high accuracy.

Challenges and Advantages of Accounting for Backbone Flexibility in Prediction of Protein–Protein Complexes

“…Only in a few cases where the subunits did not change conformation by more than a few Å in the Upside simulations did the addition of side chains with SCWRL4 26 and rescoring with SOAP-PP 23 produce a significant benefit. The CABS CG method, which also lacks explicit side chains, was also able to find some low IRMSD predictions in the top 10 scored predictions, 29 which when considered with our results indicates that moderate success in docking can be achieved without explicit side chains. However, we should note that SCWRL4 and SOAP-PP do not give a perfect reconstruction and scoring of side chains, and it is still likely that explicit side chains are required for high accuracy predictions.…”

“…Backbone Flexibility and Other MD Methods. To determine the generality of Upside's decrease in performance when backbone movement is allowed, we examined two other recent forcefield-guided methods, the CABS coarse grain model 11,12,29 and a computationally heavy all-atom explicit solvent approach. 10 Both approaches let the dynamics guide the association of the protein partners, unlike the search process used in the Upside docking pipeline to start with up to 1000 pre-docked poses.…”

“…The CABS CG model is similar to Upside but has slightly lower level detail, for example, it includes angular dependance of SC−SC interactions but single side chain states. 11,12,29 Hence, it offers an independent insight into the potential benefits of backbone flexibility for docking. The CABS model uses replica exchange with Monte Carlo moves that capture transitions on physical timescales, which they call pseudodynamics.…”

“…The partners begin in their unbound native conformations and they applied restraints individually to each partner, with strengths such that the receptor fluctuates only around 1 Å and the ligand between 2 and 12 Å. 29 This setup is overall much more flexible than the Upside unbound subunit semirigid case for the CAPRI evaluations, and the Upside case utilizes FRODOCK pre-docked starting states.…”

“…For example, we found that rigid docking performs better than the large conformational search of the CABS CG method. 29 Even with the extensive sampling enabled by Anton 2, the classical allatom MD simulations of Pan et al benefitted from imposing backbone constraints based on the bound conformation of the subunits. 10 In the case of antibody docking, biased simulations using information on antibody CDR loops and epitopes could overcome forcefield inaccuracies.…”