PRaptorX Property

PRaptorX Property predicts protein structural properties from amino acid sequence using Deep Convolutional Neural Fields (DeepCNF) to provide secondary structure (3-state and 8-state), solvent accessibility, and disorder predictions for proteins lacking close homologs in the Protein Data Bank (PDB).

Key Features:

• Deep Convolutional Neural Fields (DeepCNF): Implements a proprietary DeepCNF deep learning model to learn sequence-to-property relationships.
• Hierarchical architecture: Uses a hierarchical model architecture to capture complex sequence–structure interdependencies.
• Label correlation modeling: Accounts for correlations among adjacent property labels to improve prediction accuracy.
• Template-free prediction: Predicts properties without relying on template-based methods, enabling application to sequences with sparse profiles.
• Secondary structure (SS): Predicts both 3-state and 8-state secondary structures with approximately 84% Q3 accuracy and ~72% Q8 accuracy.
• Solvent accessibility (ACC): Estimates 3-state solvent accessibility with about 66% Q3 accuracy.
• Disorder regions (DISO): Identifies disordered regions with an AUC of approximately 0.89.

Scientific Applications:

• Analysis of proteins lacking homologs: Enables structural property prediction for proteins poorly represented in the PDB or with sparse evolutionary profiles.
• Functional annotation: Supports annotation of protein function through predicted secondary structure, accessibility, and disorder.
• Drug discovery: Provides structural property information relevant to target characterization in drug discovery workflows.
• Protein dynamics studies: Facilitates investigation of conformational flexibility and disorder in proteins.

Methodology:

Applies Deep Convolutional Neural Fields (DeepCNF) with a hierarchical architecture that models sequence–structure interdependencies and correlations among adjacent labels to produce template-free predictions from sequence or sparse sequence profiles.