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

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Structure Prediction

Background:

• Protein backbone torsion angles (Phi and Psi) are fundamental for describing local protein conformation.
• Accurate prediction of these angles is critical for understanding protein structure and function.

Purpose of the Study:

• To introduce OPUS-Refine, a general postprocessing method for improving protein structure prediction.
• To enhance the accuracy of both local (torsion angles) and global protein structural predictions.

Main Methods:

• OPUS-Refine utilizes a sampling-based approach, incorporating predictions from other methods as constraints.
• A neighbor-dependent statistical torsion angle sampling database (OPUS-TA) was developed to improve sampling efficiency.
• Contact maps from RaptorX were integrated as a global structural constraint.

Main Results:

• Refinement using OPUS-Refine increased the accuracy of Phi/Psi angles predicted by SPIDER3 and SPOT-1D.
• OPUS-Refine improved global structural accuracy (TM-score, RMSD) and local structural accuracy (Phi/Psi) compared to RaptorX online server predictions.
• The method demonstrated high efficiency, refining torsion angles in ~4s and global structures in ~30s for a 100-residue protein.

Conclusions:

• OPUS-Refine is an effective and efficient postprocessing tool for enhancing protein structure prediction accuracy.
• Its sampling-based nature allows integration with various prediction methods, offering broad applicability.
• The development of OPUS-TA further supports efficient sampling for related methods.