CRNPRED

CRNPRED predicts one-dimensional protein structural features—secondary structure (SS), contact numbers (CN), and residue-wise contact orders (RWCO)—from amino acid sequences to inform sequence–structure relationships and support three-dimensional structure prediction.

Key Features:

• Machine Learning-Based Approach: CRNPRED employs critical random networks (CRNs) that model entire amino acid sequences rather than relying on local windows.
• Prediction Targets: Generates residue-wise predictions of secondary structure (SS), contact numbers (CN), and residue-wise contact orders (RWCO) from primary sequences.
• Prediction Accuracy: Achieves average SS Q3 accuracy of ~81%, contact number correlation coefficient of 0.75, and residue-wise contact order correlation coefficient of 0.61.
• Context and Nonlinear Effects: The formulation accounts for sequence context dependence and explores nonlinear and multi-body effects beyond position-specific scoring matrices (PSSMs).
• 3D Structure Reconstruction: Predicted 1D features can be used as restraints for simulated annealing molecular dynamics to reconstruct 3D structures with coordinate RMS deviations below 4 Å from native structures.

Scientific Applications:

• 3D Structure Prediction: Provides structural restraints (SS, CN, RWCO) to aid recovery of native three-dimensional structures via simulated annealing molecular dynamics.
• Understanding Sequence–Structure Relationships: Facilitates analysis of how primary amino acid sequences encode higher-order structural features, including nonlinear and multi-body contributions.

Methodology:

Uses critical random networks (CRNs) to predict SS, CN, and RWCO from entire amino acid sequences; compares performance with linear methods based on position-specific scoring matrices (PSSMs); and demonstrates reconstruction of 3D structures using simulated annealing molecular dynamics achieving coordinate RMS deviations below 4 Å from native structures.