TM-score

TM-score quantifies the structural similarity between two protein structures to evaluate template and predicted full-length model accuracy.

Key Features:

• Protein Size-Dependent Scale: Introduces a size-dependent normalization that accounts for protein size bias and the expected similarity of random protein structure pairs.
• Comprehensive Residue Pair Evaluation: Evaluates all residue pairs in an alignment rather than applying fixed distance cutoffs, providing a continuous assessment of structural similarity.
• Correlation to Model Quality: Demonstrates strong correlation with full-length model quality, validated on a benchmark of 1,489 small to medium-sized proteins using threading programs PROSPECTOR_3 and modeling tools MODELLER and TASSER.
• CASP5 Assessment: Was applied to assess 'new fold' targets in CASP5, producing results that closely aligned with human-expert visual assessments.

Scientific Applications:

• Quality Assessment: Provides a quantitative metric for comparing predicted protein structures to native structures.
• Modeling and Threading: Serves as an evaluation metric in protein modeling and threading workflows to compare templates and predicted models.
• Research and Development: Used to benchmark and refine computational structure-prediction methods in structural biology.

Methodology:

Computes a size-normalized scoring function that evaluates all residue pairs without fixed distance cutoffs, weights close matches more than distant ones, and builds on concepts from GDT and MaxSub to achieve greater sensitivity than RMSD.