OPUS-Refine: A Fast Sampling-Based Framework for Refining Protein Backbone Torsion Angles and Global Conformation.

Abstract

Protein backbone torsion angles (Phi and Psi) are crucial for protein local conformation description. In this paper, we propose a general postprocessing method for all prediction methods, namely, OPUS-Refine, which may contribute to the field in a different way. OPUS-Refine is a sampling-based method, therefore, the results of other prediction methods can be used as its constraints. After OPUS-Refine refinement, for instance, the accuracy of Phi/Psi predicted by SPIDER3 and SPOT-1D are both increased. In addition, to facilitate the sampling efficiency, we construct a neighbor-dependent statistical torsion angles sampling database, namely, OPUS-TA, which may be useful for other sampling-based methods. Furthermore, we also introduce the contact map predicted by RaptorX to OPUS-Refine as a global structural constraint. After refinement, compared to the predicted structures obtained from RaptorX online server, the accuracy of both global structural configurations (measured by TM-score and RMSD) and local structural configurations (measured by Phi/Psi) results are improved. OPUS-Refine is a highly efficient framework, it takes only about 4 s to refine the torsion angles and 30 s to refine the global structure of a protein with 100 residues in length on a typical desktop personal computer. Therefore, the sampling-based feature and the efficiency of OPUS-Refine offer greater potentiality for it to take advantage of any other method to achieve better performance.