QDeep

QDeep: Distance-Based Protein Model Quality Estimation Using Deep Residual Networks

QDeep performs protein model quality estimation by integrating inter-residue distance information with stacked deep residual neural networks (ResNets) to predict residue-level errors and compute global quality scores for protein structural models.

Key Features:

• Distance-Based Modeling: Incorporates inter-residue distance information as input features for model quality estimation.
• Stacked Deep Residual Neural Networks (ResNets): Utilizes stacked deep ResNets to perform residue-level ensemble error classification at multiple predefined error thresholds.
• Ensemble Error Classification: Classifies residue-level errors across multiple thresholds and integrates predictions to generate comprehensive model-level quality estimates.

Scientific Applications:

• Protein Structure Validation: Evaluates and ranks predicted protein structural models in structural biology and bioinformatics workflows, outperforming ProQ2, ProQ3, ProQ3D, ProQ4, 3DCNN, MESHI, and VoroMQA across independent test datasets.

Methodology:

QDeep applies stacked deep ResNets to inter-residue distance–derived features to perform residue-level error classification at multiple predefined thresholds, then aggregates ensemble predictions to compute global protein model quality scores.