DAEMDA: A Method with Dual-Channel Attention Encoding for miRNA-Disease Association Prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Aberrant microRNA (miRNA) expression is linked to complex human diseases.
• Identifying miRNA-disease associations (MDA) is crucial for understanding disease pathogenesis and developing therapies.
• Existing graph neural network (GNN) methods for MDA prediction often neglect global network features.

Purpose of the Study:

• To develop an advanced computational method, DAEMDA, for improved miRNA-disease association prediction.
• To address the limitation of existing GNN models in exploiting global network information for high-quality embedding representations.

Main Methods:

• Construction of similarity and heterogeneous networks based on miRNA-disease association data and analogous information.
• Implementation of a parallel dual-channel feature encoder to capture global network information and generate diverse embedding representations.
• Utilization of a neural network classifier to merge dual-channel embeddings for predicting miRNA-disease associations.

Main Results:

• DAEMDA generates high-quality embedded representations by effectively leveraging global network properties.
• The method significantly improves the accuracy of miRNA-disease association prediction.
• Validation through five-fold cross-validation and case studies on the HMDD v3.2 database confirms efficacy.

Conclusions:

• DAEMDA offers a superior approach to predicting miRNA-disease associations compared to existing methods.
• The model's ability to integrate global network features enhances biomarker discovery and understanding of disease mechanisms.