RaptorX-Angle

RaptorX-Angle predicts real-valued protein dihedral angles from amino acid sequences to characterize local protein conformation and narrow conformational space for tertiary structure prediction.

Key Features:

• Novel prediction approach: Integrates clustering techniques with deep learning algorithms to predict real-valued dihedral angles directly from protein sequences.
• Performance and benchmarking: Evaluated on datasets including a subset of PDB25 and targets from recent CASP experiments, showing higher Pearson Correlation Coefficient (PCC) and lower Mean Absolute Error (MAE) compared to SPIDER2.
• Error estimation: Provides estimated bounds on predictions whose relationship to real prediction errors is approximately linear, allowing approximation of prediction accuracy.

Scientific Applications:

• Protein structure prediction: Reduces local conformational ambiguity to improve tertiary structure modeling.
• Functional and folding studies: Supplies local dihedral angle information useful for analyzing protein folding mechanisms, interactions, and function.
• Drug discovery and interaction analysis: Offers structural constraints that can inform modeling of ligand binding and protein–protein interactions.

Methodology:

Combines clustering techniques with deep learning algorithms to directly predict real-valued dihedral angles from protein sequences.