HGSMDA: miRNA-Disease Association Prediction Based on HyperGCN and Sørensen-Dice Loss

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genomics

Background:

• Identifying microRNA (miRNA)-disease associations is vital for disease prevention, diagnosis, and treatment.
• Experimental methods for inferring these associations are costly and inefficient.
• Current predictive models, often based on Graph Convolutional Networks (GCNs), have limitations in aggregating information from multiple nodes.

Purpose of the Study:

• To propose a novel, accurate, and efficient computational approach for predicting miRNA-disease associations.
• To overcome the limitations of traditional GCNs in capturing complex relationships.
• To introduce a new framework, HGSMDA, leveraging HyperGCN and Sørensen-Dice loss.

Main Methods:

• Constructed multiple similarity networks for miRNAs and diseases.
• Employed GCNs for feature extraction from diverse perspectives.
• Utilized HyperGCN to build a miRNA-disease heteromorphic hypergraph.
• Trained GCN on the hypergraph to aggregate information.
• Applied Sørensen-Dice loss for evaluating prediction accuracy.

Main Results:

• The proposed HGSMDA method demonstrated superior efficacy compared to existing methodologies.
• Extensive experiments on the Human MicroRNA Disease Database (HMDD v3.2) validated the model's performance.
• A case study on colon cancer further corroborated the predictive power of HGSMDA.

Conclusions:

• HGSMDA provides a dependable and valid framework for predicting miRNA-disease associations.
• The novel approach offers a promising avenue for investigating complex miRNA-disease relationships.
• This method enhances the efficiency and accuracy of identifying potential biomarkers for diseases.