Prediction of circRNA-Disease Associations Based on Graph Isomorphism Networks and Graph Sampling Aggregation.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Understanding circular RNA (circRNA) and disease relationships is vital for disease mechanism research.
• Experimental methods for identifying circRNA-disease associations are time-consuming and labor-intensive.
• Existing computational methods have limitations in deep feature extraction.

Purpose of the Study:

• To develop an innovative computational framework, GINSACDA, for predicting unknown circRNA-disease associations.
• To overcome limitations of existing methods by integrating diverse feature types and advanced network architectures.

Main Methods:

• GINSACDA computes Gaussian interactive profile kernel (GIP) similarity and functional/semantic similarity for circRNAs and diseases as global features.
• Local features are extracted from seven-hop subgraphs.
• Fused global and local features are processed by Graph Isomorphism Network (GIN) and Graph Sampling Aggregation (GraphSAGE) for deep feature extraction.
• A fully connected layer calculates prediction scores.

Main Results:

• GINSACDA demonstrated superior performance compared to five state-of-the-art models in five-fold cross-validation on two datasets.
• Case studies on hepatocellular carcinoma and breast cancer validated the model's predictive power.

Conclusions:

• GINSACDA offers a powerful and accurate computational approach for predicting circRNA-disease associations.
• The framework's ability to extract deep features from integrated data enhances understanding of disease mechanisms.