Scoring function for automated assessment of protein structure template quality

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Assessing protein structure quality is crucial for understanding biological function.
• Existing scoring functions like GDT and MaxSub have limitations, including protein size dependence and reliance on specific distance cutoffs.

Purpose of the Study:

• To develop and validate a novel scoring function, the Template Modeling score (TM-score), for evaluating protein structure templates and predicted models.
• To address the limitations of existing scoring methods by accounting for protein size and evaluating all residue pairs.

Main Methods:

• Developed the TM-score by extending Global Distance Test (GDT) and MaxSub approaches.
• Incorporated a protein size-dependent scale to normalize scores.
• Evaluated all residue pairs in alignments/models, not just those below specific cutoffs.
• Constructed a benchmark set of 1489 protein structure templates and models using PROSPECTOR_3, MODELLER, and TASSER.
• Compared TM-score performance against GDT and MaxSub using this benchmark set.
• Applied TM-score to assess 'new fold' targets in the CASP5 experiment.

Main Results:

• The TM-score demonstrated a stronger correlation with the quality of final full-length protein models compared to GDT and MaxSub.
• TM-score performance on CASP5 'new fold' targets closely matched human expert assessments.
• The developed scoring function effectively eliminates inherent protein size dependence and accounts for random structure pairs.

Conclusions:

• The TM-score is a robust and reliable metric for assessing protein structure prediction quality.
• TM-score offers a valuable complement to existing automated and human-based assessment methods.
• The TM-score program is publicly available for use in bioinformatics research.