Evaluation of model refinement in CASP13.

Area of Science:

• Computational biology
• Structural biology
• Protein structure prediction

Background:

• The Critical Assessment of Structure Prediction (CASP) is a community-wide experiment to assess protein structure prediction methods.
• Model refinement is a crucial step in protein structure prediction, aiming to improve the quality of initial models.

Purpose of the Study:

• To assess the performance of different groups in the model refinement category of CASP13.
• To identify common strategies employed by successful refinement methods.
• To propose improved metrics for evaluating protein model quality.

Main Methods:

• Analysis of results from the CASP13 model refinement category.
• Application of the CASP12 ranking formula to CASP13 data.
• Evaluation of traditional and alternative metrics for model quality assessment, including main-chain and side-chain torsion angles, and utility for molecular replacement.
• Consideration of coordinate error estimates.

Main Results:

• Only 7 out of 32 participating groups consistently improved starting models; the majority degraded them on average.
• Successful refinement approaches often rely on physics-based force fields and molecular dynamics with restraints.
• Traditional CASP metrics may not sufficiently discriminate between good and very good models.

Conclusions:

• Protein model refinement remains a challenging task, with limited success among participants.
• Physics-based approaches and molecular dynamics are key components of effective refinement.
• Introducing new evaluation metrics, such as torsion angle comparisons and error estimate accuracy, could enhance the assessment of protein model quality.