Improving low-accuracy protein structures using enhanced sampling techniques.

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Protein structure prediction and refinement are critical for understanding biological function.
• Accurate protein models are essential for drug discovery and molecular biology.
• Enhanced sampling techniques can improve the efficiency and accuracy of molecular simulations.

Purpose of the Study:

• To improve the accuracy of protein structure refinement using enhanced sampling and blind selection methods.
• To evaluate the performance of a combined PCST-SBM approach with a novel model selection strategy.
• To assess the potential for improving low-accuracy protein structures with current computational methods.

Main Methods:

• Utilized parallel continuous simulated tempering (PCST) combined with structure-based model (SBM) restraints.
• Applied the methods to refine 23 protein targets from CASP10 and CASP12.
• Developed and implemented a novel blind model selection technique for long simulation trajectories.

Main Results:

• The combined PCST-SBM and blind selection method produced refined models superior to initial models.
• In the Top-1 group, 7 out of 23 targets showed improved models compared to Critical Assessment of Structure Prediction (CASP) participants.
• In the Top-5 group, 10 out of 23 targets achieved better model accuracy.

Conclusions:

• Enhanced sampling plays a vital role in protein structure prediction and refinement.
• Significant improvements in low-accuracy protein structures are achievable with existing force fields.
• The developed blind selection method effectively identifies high-quality models from complex simulation data.