THGNCDA: circRNA-disease association prediction based on triple heterogeneous graph network

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics
• Genomics and Genetics

Background:

• Circular RNAs (circRNAs) are noncoding RNA molecules with a closed circular structure implicated in various diseases.
• circRNAs show potential as biomarkers for clinical diagnosis and disease treatment.
• Traditional experimental methods for identifying circRNA-disease associations are time-consuming and labor-intensive.

Purpose of the Study:

• To develop a novel computational method, THGNCDA, for predicting circRNA-disease associations.
• To incorporate topological information and miRNA interactions into circRNA-disease association prediction.
• To provide a more efficient and accurate approach compared to existing machine learning methods.

Main Methods:

• Employed a graph neural network with an attention mechanism to learn neighbor importance for circRNA-disease pairs.
• Utilized a multilayer convolutional neural network to explore relationships based on circRNA and disease attributes.
• Integrated miRNA information during the embedding calculation process.

Main Results:

• THGNCDA demonstrated superior performance compared to state-of-the-art (SOTA) methods in predicting circRNA-disease associations.
• The proposed method achieved a better recall rate, indicating improved identification of true associations.
• Ablation studies confirmed the significance of the attention mechanism in the model's performance.

Conclusions:

• THGNCDA offers an effective and efficient computational approach for predicting circRNA-disease associations.
• The method's ability to discover known relationships, as shown in case studies, highlights its potential for identifying novel associations.
• Incorporating topological and miRNA information enhances the accuracy of circRNA-disease association prediction.