Predicting CircRNA disease associations using novel node classification and link prediction models on Graph...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (CircRNAs) are increasingly recognized for their roles in complex human diseases.
• CircRNAs hold potential as biomarkers for disease diagnosis and therapeutic targets.
• Experimental verification of CircRNA-disease associations is limited by scale, time, and cost.

Purpose of the Study:

• To develop effective computational methods for predicting CircRNA-disease associations.
• To leverage deep learning and graph-based approaches for enhanced prediction accuracy.
• To provide a computational tool to guide experimental validation efforts.

Main Methods:

• Integration of multi-source similarity information into a CircRNA-disease association network.
• Application of Graph Convolutional Networks (GCN) for association prediction.
• Development of two GCN-based models: GCN for Node Classification (GCN-NC) and GCN for Link Prediction (GCN-LP).

Main Results:

• The proposed GCN-based models demonstrate promising predictive performance.
• The novel models outperform existing shallow learning-based prediction methods.
• Experimental validation through a case study confirmed predicted associations with published literature and gene-gene interaction networks.

Conclusions:

• GCN-based models offer a powerful deep learning approach for CircRNA-disease association prediction.
• These computational models can significantly accelerate the identification of potential CircRNA biomarkers and therapeutic targets.
• The study highlights the potential of integrating network structure and deep learning for biological association prediction.