GCSENet

GCSENet predicts microRNA (miRNA)–disease associations by integrating heterogeneous disease, gene, and miRNA data using graph convolutional networks (GCNs), convolutional neural networks (CNNs), and squeeze-and-excitation (SENet) modules to prioritize informative features.

Key Features:

• Heterogeneous graph integration: Integrates diseases, genes, and miRNAs into a heterogeneous graph representation.
• Graph Convolutional Networks (GCNs): Uses GCNs to capture topological and feature representations from the heterogeneous graph.
• Feature weighting: Assigns feature weights to gene–miRNA and disease–gene relationships to model differential impacts of genes on miRNAs and disease categories.
• Convolutional processing (CNNs): Applies CNNs to graph-derived features for additional hierarchical feature extraction.
• Squeeze-and-Excitation (SENet) attention: Incorporates SENet squeeze-and-excitation blocks to evaluate and recalibrate the importance of each feature channel.
• Evaluation: Validated by 10-fold cross-validation reporting AUROC 95.02% and AUPR 95.55%, with reported superior performance versus other state-of-the-art methods.

Scientific Applications:

• miRNA–disease association prediction: Identifies potential associations between microRNAs and human diseases.
• Prioritization of pathogenic factors: Helps uncover and prioritize miRNAs and gene interactions that may contribute to disease pathogenesis.
• Support for drug development and personalized medicine: Informs discovery and prioritization efforts relevant to therapeutic development and personalized treatment strategies.

Methodology:

Construct a heterogeneous graph of diseases, genes, and miRNAs; apply graph convolutional networks (GCNs) to extract node and relational features; assign feature weights to gene–miRNA and disease–gene relationships; process weighted features with convolutional neural networks (CNNs) and squeeze-and-excitation (SENet) blocks for channel-wise attention; and evaluate performance using 10-fold cross-validation with AUROC and AUPR metrics.