DeepHomo

DeepHomo predicts inter-protein residue-residue contacts within homo-oligomeric and homo-dimeric protein complexes to support structural characterization of protein-protein interfaces.

Key Features:

• Deep learning integration: Employs deep learning methodologies to combine multiple data types for contact prediction.
• Intra-Protein Distance Maps: Incorporates intra-protein distance maps to provide spatial information about residues within monomers.
• Inter-Protein Docking Patterns: Utilizes inter-protein docking patterns to inform likely association modes at interfaces.
• Evolutionary Coupling and Sequence Conservation: Leverages evolutionary coupling and sequence conservation signals to identify functionally important residue pairs.
• Physico-Chemical Monomer Information: Integrates physico-chemical properties of monomers to refine contact predictions.
• Joint MSA and homolog limitation handling: Addresses the limited number of known homologous protein-protein interactions and the difficulty of generating joint multiple sequence alignments for interacting proteins.
• Performance: Achieves over 60% precision for top predicted contacts compared to direct-coupling analysis and other machine learning methods.
• Validation: Validated on experimentally determined structures and realistic targets from CASP and CAPRI.
• Docking integration: Predicted inter-chain contacts can be integrated into protein-protein docking to substantially improve docking accuracy on benchmark homo-dimeric targets.

Scientific Applications:

• Structural characterization: Determining residue-residue interfaces in homo-oligomeric and homo-dimeric protein complexes.
• Protein-protein docking: Guiding and improving docking predictions via incorporation of predicted inter-chain contacts.
• Mechanistic insight: Informing studies of molecular mechanisms of protein-protein interactions through contact-level information.
• Therapeutic targeting: Supporting drug development efforts that target protein-protein interfaces by identifying critical contact residues.
• Benchmarking and evaluation: Providing validated contact predictions for assessment in CASP- and CAPRI-style challenges.

Methodology:

Uses deep learning to integrate intra-protein distance maps, inter-protein docking patterns, evolutionary coupling and sequence conservation, and physico-chemical information of monomers to predict inter-protein residue-residue contacts.