MAGCDA: A Multi-Hop Attention Graph Neural Networks Method for CircRNA-Disease Association Prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Circular RNAs (circRNAs) play a crucial role in human diseases.
• Disease-associated circRNAs are valuable biomarkers and therapeutic targets.
• Current computational methods for circRNA-disease association (CDA) prediction lack contextual information, limiting performance.

Purpose of the Study:

• To develop an advanced computational approach for predicting circRNA-disease associations (CDAs).
• To improve the accuracy of CDA prediction by incorporating biological network context.
• To identify novel disease-associated circRNAs for diagnostic and therapeutic applications.

Main Methods:

• Constructed a multi-source attribute heterogeneous network of circRNAs and diseases.
• Employed a multi-hop attention graph neural network (MAGCDA) to aggregate node context information.
• Utilized a random forest classifier for final CDA inference.

Main Results:

• MAGCDA achieved high prediction accuracies (92.58%, 91.42%, 83.46%, 91.12%) on four gold standard datasets.
• Demonstrated superior performance compared to existing models in ablation experiments and comparative analyses.
• Identified 18 and 17 potential disease-associated circRNAs for breast cancer and Alzheimer's disease, respectively, in case studies.

Conclusions:

• MAGCDA effectively infers potential circRNA-disease associations by leveraging network context.
• The method provides a robust tool for exploring novel disease-associated circRNAs.
• Results offer a theoretical foundation for disease diagnosis and treatment strategies involving circRNAs.