QDistance

QDistance assesses protein model quality by leveraging inter-residue distance predictions from the deep learning method trRosetta to estimate global and local accuracy of structural models.

Key Features:

• Input handling: Accepts either single models or multiple models as input.
• Distance-based features: Computes features that measure concordance between trRosetta-predicted inter-residue distances and distances derived from structural models.
• Integration with QA metrics: Combines distance-based features with several widely recognized QA metrics.
• Global scoring: Uses a linear regression framework to integrate features and produce global quality scores.
• Local quality assessment: Performs local assessment by comparative analysis against a set of reference models.
• Reference selection: For multiple-model inputs, selects reference models based on predicted global quality, and for single-model inputs, generates reference models using trRosetta predictions.
• Unreliable region detection: Identifies unreliable local regions through distance-based comparisons.

Scientific Applications:

• Global model ranking: Estimates global quality scores to rank or select the most accurate models without access to native structures.
• Local accuracy estimation: Provides local quality estimates and identification of unreliable local regions within models.
• Benchmarking: Demonstrated competitive performance on CASP13 and CAMEO structure model benchmarks.
• Blind evaluation: Achieved robust results in CASP14 blind testing, ranking among top predictors and within the top three for local quality assessment.

Methodology:

Uses trRosetta deep learning predictions of inter-residue distances; computes distance-based concordance features between predicted and model-derived distances; integrates these features with several QA metrics via linear regression to produce global scores; performs local assessment by comparative analysis against reference models selected or generated as described.