CEthreader

CEthreader integrates predicted residue-residue contact maps with sequence profile alignments to improve protein fold recognition and template selection for atomic-level structure prediction.

Key Features:

• Contact-map and profile integration: Integrates contact-map predictions with sequence profile alignments to couple spatial restraints with alignment scoring.
• Contact prediction model: Predicts residue-residue contacts by combining evolutionary precision matrices with deep residual convolutional neural networks (CNNs).
• Template identification: Uses predicted contact maps together with sequence profiles to identify structural templates from the Protein Data Bank (PDB).
• Enhanced fold recognition: Couples predicted contact maps with profile alignments to improve recognition of global protein folds for distant-homology proteins.
• Benchmark performance: On two independent benchmark sets of 1,153 non-homologous targets, detected 176% more correct templates (TM-score > 0.5) for "Hard" targets lacking homologous templates, 36% more than contact-based methods, and 114% more when excluding structures from the same SCOPe Superfamily.
• Relevance to atomic-level prediction: Enhances template-based modeling (TBM) accuracy for atomic-level protein structure prediction.

Scientific Applications:

• Template-based modeling (TBM): Improves selection of PDB templates and TBM accuracy for distant-homology proteins.
• Protein fold recognition: Enhances recognition of global protein folds when homologous relationships are weak or absent.
• Functional annotation and ligand design: Supports atomic-level structure predictions used for annotating biological function and guiding compound design that modulates protein function.

Methodology:

Predicts residue-residue contacts by combining evolutionary precision matrices with deep residual CNNs, integrates those contact-map predictions with sequence profile alignments, and uses the combined information to identify structural templates from the PDB.