CRPGCN

CRPGCN predicts associations between circular RNAs (circRNAs) and diseases using a graph convolutional network (GCN)-based approach to compute similarity scores and identify potential biomarkers.

Key Features:

• Graph Convolutional Network (GCN): Employs a GCN to learn feature representations from network-structured data and compute similarity scores between circRNAs and diseases.
• Random Walk with Restart (RWR): Integrates RWR to enhance node similarity by iteratively exploring neighboring nodes and refining connectivity patterns.
• Principal Component Analysis (PCA): Applies PCA for dimensionality reduction and extraction of significant features prior to learning.
• Heterogeneous Network Construction: Constructs heterogeneous adjacency and feature matrices by combining an adjacency matrix, similarity matrix, and feature matrix to represent circRNA–disease relationships.

Scientific Applications:

• Biomarker Discovery: Predicts potential circRNA–disease associations to aid identification of novel diagnostic and prognostic biomarkers.
• Mechanistic Insights: Elucidates relationships between circRNAs and diseases to inform hypotheses about underlying disease mechanisms.
• Cost-Effective Research: Provides a computational approach that complements experimental methods for studying circRNA–disease relationships.

Methodology:

Constructs a heterogeneous network by combining adjacency, similarity, and feature matrices into heterogeneous adjacency and feature matrices; enhances node similarity using Random Walk with Restart (RWR); performs dimensionality reduction and feature extraction with PCA; applies a GCN to the heterogeneous matrices to learn representations and compute final similarity scores; evaluates performance via 2-fold, 5-fold, and 10-fold cross-validation with reported AUCs of 0.9490, 0.9720, and 0.9722, respectively.