MRGBMDAT: A Multi-Relational Graph Encoder Network with Bilinear Fusion for miRNA-Disease Association Type Prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial regulators of biological processes, and their dysregulation is linked to various diseases.
• Accurate prediction of miRNA-disease associations is vital for understanding disease mechanisms and advancing precision medicine.
• Existing computational methods often struggle with integrating complex data structures and handling imbalanced datasets.

Purpose of the Study:

• To propose MRGBMDAT, a multi-relational graph encoder network with bilinear fusion for enhanced miRNA-disease association type prediction.
• To address limitations of existing methods by better integrating local semantic dependencies and global topological structures.
• To mitigate the class imbalance problem in miRNA-disease association prediction.

Main Methods:

• MRGBMDAT employs a multi-relational graph convolution module with bidirectional cross-attention for global structure analysis.
• A local subgraph sampling module captures local semantic dependencies.
• Bilinear fusion with element-wise attention models interactions, and an iterative negative sampling strategy addresses class imbalance.

Main Results:

• MRGBMDAT significantly outperforms five state-of-the-art methods on the HMDD v3.2 dataset.
• The model demonstrates strong discriminative power and generalization capability in predicting miRNA-disease association types.
• Experimental results validate the effectiveness of the proposed integrated approach.

Conclusions:

• MRGBMDAT offers a superior computational approach for predicting miRNA-disease association types.
• The method's ability to integrate diverse data aspects and handle imbalanced data makes it a valuable tool for biomedical research.
• This work contributes to advancing precision medicine through improved understanding of miRNA-disease relationships.