Estimating Residue-Specific Accuracies of Protein Structure Models Using Molecular Dynamics Simulations

Abstract

Estimating the accuracy of a structure model is very crucial to promote the usefulness of protein structure prediction methods. Currently, a vast majority of successful model quality assessment (or model accuracy estimation, MAE) methods are knowledge-based. Based on molecular dynamics (MD) simulation with a recently developed residue-specific force field (RSFF2), we develop a method for absolute MAE at per-residue level. Using a training set of 31 models and a test set of 24 models from different proteins, the MAE performance of our MD-based method can reach or even exceed the state-of-the-art single-model MAE method within a short simulation time (less than one nanosecond). In addition, a simple combination of knowledge-based method with the MD-based method can obtain more accurate MAE than any of the constituent methods.