BND

BND refines residue contact maps derived from co-evolution data to distinguish direct from indirect residue dependencies and thereby improve protein structure prediction.

Key Features:

• Balanced Network Deconvolution (BND): Identifies an optimized dependency matrix without constraints on the eigenvalue range within applied network systems.
• Direct vs indirect distinction: Separates direct residue dependencies from indirect and transitive contacts in co-evolution-based predictions.
• Contact prediction improvement: Improves medium-range contact prediction by 55.59% and long-range contact prediction by 47.68% at an L/5 cutoff on CASP9, CASP10, and PSICOV benchmark datasets.
• Computational cost: Achieves the reported improvements without incurring additional computational cost.
• Statistical validation: Enhancements were confirmed significant (P < 5.93 × 10^(-3)) by Student's t-test.
• General refinement method: Functions as a general contact refinement approach applicable to co-evolution-based contact predictions.
• Dependence on homologs: Effectiveness is contingent on the availability of homologous sequences in sequence databases.

Scientific Applications:

• Protein structure prediction: Refines co-evolution-derived residue contact maps to improve three-dimensional protein structure modeling.
• Ab initio structure prediction (CASP): Supplies improved contact inputs for ab initio structure prediction in CASP experiments, subject to homologous sequence availability.
• Medium- and long-range contact analysis: Enhances medium- and long-range contact predictions used in structural modeling and fold recognition.

Methodology:

Based on the approach of Feizi et al., BND identifies an optimized dependency matrix without eigenvalue-range constraints, distinguishes direct and indirect residue dependencies, evaluates contact prediction performance on CASP9, CASP10, and PSICOV at an L/5 cutoff, and assesses significance using Student's t-test (P < 5.93 × 10^(-3)).