ProQ

ProQ predicts the quality of protein structure models to distinguish native or near-native conformations and evaluate model accuracy.

Key Features:

• Neural-network scoring: Uses a neural-network-based approach to predict model quality.
• Structural-fragment detection: Identifies native or near-native models and models with limited structural similarity by detecting significant fragments similar to the native conformation.
• Atom-atom contact features: Extracts structural features including the frequency of atom-atom contacts within protein models.
• Quality metrics: Predicts model quality using established metrics such as LGscore and MaxSub.
• Integration with fold recognition: Integrates with Pcons fold recognition predictors through the Pmodeller system to refine predictions and eliminate high-scoring incorrect models.
• Benchmark performance: Demonstrated superior performance in identifying correct models across test sets and in CASP5 and LiveBench-6.

Scientific Applications:

• Model quality assessment: Discriminates accurate protein models from less precise ones for protein structure prediction workflows.
• Fold recognition enhancement: Improves fold recognition by detecting models containing significant native-like fragments beyond global similarity measures.
• Post-processing of fold predictors: Refines Pcons predictions via Pmodeller integration to increase specificity by removing high-scoring incorrect models.
• Evaluation using LGscore and MaxSub: Serves to evaluate and rank predicted structures using LGscore or MaxSub.

Methodology:

Applies a neural-network-based model that extracts structural features such as atom-atom contact frequency and predicts model quality using LGscore or MaxSub, and can be integrated with Pcons via Pmodeller to refine predictions.