A multi-source similarity fusion method based on hypergraph convolutional networks and graph transformers for...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying miRNA-disease associations (MDA) is crucial for drug development but traditional methods are costly and time-consuming.
• Existing computational MDA prediction methods suffer from low accuracy, poor generalization, and reliance on single data sources.
• There is a need for advanced computational approaches to efficiently and accurately predict MDA.

Purpose of the Study:

• To propose a novel computational method, HGC-GraphT, for predicting miRNA-disease associations.
• To address limitations of existing methods by integrating multi-source biological data and employing hypergraph convolutional networks.
• To improve the accuracy and reliability of MDA predictions.

Main Methods:

• HGC-GraphT integrates multi-source similarity information including drug-disease, protein, circRNA, and LncRNA data.
• A hypergraph is constructed to mine complex association information between multiple biological nodes.
• Hypergraph convolutional neural networks and a biological entity graph are utilized for comprehensive feature extraction.

Main Results:

• HGC-GraphT demonstrated superior prediction performance compared to five other methods across multiple metrics (AUC, AUPR, ACC, F1, recall, precision).
• Experimental validation using the HMDD v3.2 database confirmed the effectiveness of the proposed method.
• Case studies successfully identified miRNAs associated with colon cancer and gastrointestinal tumors.

Conclusions:

• The HGC-GraphT method provides an effective computational approach for predicting miRNA-disease associations.
• The method offers a reliable auxiliary tool for advancing biomedical research and understanding disease mechanisms.
• Integrating diverse biological data through hypergraph networks enhances the predictive power for MDA.