Protein-Protein Complex Structure Prediction using the Solution Theory in the Energy Representation

Abstract

Proteins conduct their functions through interactions with other molecules. Thus, accurate prediction of protein complex is a key to understand functional mechanisms. We have developed a method to evaluate binding free energy differences of complex models generated by docking prediction using the all-atom molecular dynamics simulation and solution theory in the energy representation. By evaluating binding free energy difference with the method, we previously show that “near-native” models similar to crystal structure are successfully selected as the lowest energy structures in two protein-protein complexes, bovine trypsin with CMTI-1 squash inhibitor and RNase SA with barstar [Takemura, J. Chem. Phys. 2012]. The method requires relatively short MD simulations (2 ns) and can calculate binding free energy difference among several hundreds of complex models. As a protein-protein complex structure prediction method, we combined the evaluation method with CyClus [Omori, Proteins, 2013] that performs fast clustering/reranking using a cylindrical approximation of interface and improves the results of the rigid-body docking.The initial complex models were prepared using the protein-protein rigid body docking program, ZDOCK. The generated complex models were then clustered and reranked using CyClus. Free energy analysis for top 100 or 300 complex models returned by CyClus were conducted using conformational energies, solute entropies, and solvation free energies calculated with the solution theory in the energy representation. Our analysis improved the results of CyClus and the complex models similar to the native structure have the lowest binding free energies, suggesting that this procedure is effective in protein-protein complex structure predictions. We further improved the procedure by placing interface waters into protein-protein interface before free energy evaluations.