Comparative analysis of methods for evaluation of protein models against native structures

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• Accurate assessment of protein models is crucial for advancing protein structure prediction.
• Existing evaluation methods lack comprehensive, unbiased comparison.

Purpose of the Study:

• To comparatively analyze popular protein model assessment methods.
• To reveal the relative strengths and weaknesses of these scoring metrics.
• To provide guidance for selecting appropriate assessment scores.

Main Methods:

• Comparative analysis of RMSD, TM-score, GDT, QCS, CAD-score, LDDT, SphereGrinder, and RPF.
• Utilized a large, diverse model set from CASP10-12.
• Examined score distributions, correlations, and model selection differences.
• Assessed score variations based on structural properties like stereochemistry, hydrogen bonds, packing, and missing residues.

Main Results:

• Identified general differences in score distributions, correspondence, and correlations.
• Demonstrated how different scores impact the selection of top-ranked models.
• Showcased score sensitivity to various structural features of protein models.

Conclusions:

• Established a basis for informed selection of protein model assessment scores.
• Highlighted the importance of considering structural properties in evaluation.
• Facilitated optimal choice of scoring metrics for specific protein structure prediction challenges.