CDHGNN

CDHGNN predicts potential associations between circular RNAs (circRNAs) and diseases by integrating multi-source pathogenesis data with heterogeneous graph neural networks to identify disease-associated circRNAs and their regulatory relationships.

Key Features:

• Multi-Source Data Aggregation: Integrates multi-source pathogenesis data to mitigate data sparsity in circRNA-disease association studies.
• Heterogeneous Graph Neural Networks (GNNs): Uses heterogeneous GNNs to model and preserve the inherent heterogeneity of multi-source biological data.
• Edge-Weighted Graph Attention Network: Implements an edge-weighted graph attention mechanism to assign differing importance to edge types and capture association probabilities between nodes.
• Meta-Path Learning: Employs contextual meta-path learning with attention weights over diverse edge types to learn relationships within the heterogeneous network.
• Network Construction: Constructs four networks—circRNA network, microRNA network, disease network, and a heterogeneous network—based on integrated data sources.
• Node Feature Extraction: Extracts node features that reflect association probabilities using the edge-weighted attention model.

Scientific Applications:

• Biomarker Identification: Identifies disease-associated circRNAs that can serve as biomarkers for diagnosis.
• Therapeutic Target Discovery: Supports discovery of circRNAs as potential therapeutic targets by predicting disease associations.
• Regulatory Network Investigation: Enables investigation of biomolecular regulatory relationships involving circRNAs, microRNAs, and diseases at the systems level.
• Pathogenesis Studies: Facilitates system-level understanding of disease pathogenesis through predicted molecular associations.

Methodology:

Constructs circRNA, microRNA, disease, and heterogeneous networks from multi-source data; integrates these networks using an edge-weighted graph attention model to extract node features reflecting association probabilities; and applies heterogeneous neural networks and contextual meta-path learning with attention weights to infer potential circRNA-disease associations.