HIGH-PPI

HIGH-PPI predicts protein-protein interactions (PPIs) and localizes critical binding and catalytic sites by combining double-viewed hierarchical graph learning across the human interactome with internal protein graphs represented by chemically relevant descriptors.

Key Features:

• Double-viewed hierarchical graph learning: Constructs a hierarchical graph that couples an external top view of proteins in the PPI network with an internal bottom view for each protein.
• Top-view PPI representation: Represents nodes as proteins within the human interactome to capture macroscopic interaction patterns.
• Bottom-view internal protein graphs: Models each protein internally as a detailed graph using chemically relevant descriptors instead of traditional protein sequences.
• Chemically relevant descriptors: Encodes structure-function relationships of proteins via chemical descriptors to inform interaction prediction.
• Integration of macroscopic and microscopic perspectives: Integrates protein-level internal structure and network-level interactions for joint analysis.
• Binding and catalytic site identification: Identifies critical binding and catalytic sites to interpret modes of action underlying PPIs.
• Domain-knowledge-driven interpretability: Leverages domain knowledge to provide interpretable predictions of PPIs and molecular details.
• Accuracy and robustness: Employs the dual-view hierarchical approach to achieve reported high accuracy and robustness in PPI prediction.

Scientific Applications:

• PPI prediction: Predicts protein-protein interactions across the human interactome.
• Molecular mechanism elucidation: Elucidates molecular details and modes of action by localizing binding and catalytic sites.
• Structure-function analysis: Analyzes protein structure-function relationships using chemically relevant descriptors at the protein level.
• Interactome-level studies: Supports analysis of macroscopic interactome organization informed by internal protein structure.

Methodology:

Constructs a hierarchical PPI graph (top view), models each protein as an internal graph with chemically relevant descriptors (bottom view), and integrates these dual graph perspectives using hierarchical graph learning.