iDRKAN: Interpretable miRNA-Disease Association Prediction Based on Dual-Graph Representation Learning and...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNA-disease association (MDA) identification is crucial for biomedical research and clinical applications.
• Existing computational methods often struggle with complex network semantics and lack transparency due to their "black-box" nature.
• There is a need for interpretable and accurate methods for MDA prediction.

Purpose of the Study:

• To propose an interpretable microRNA-disease association prediction method (iDRKAN).
• To enhance the capture of deep semantic information in complex biological networks.
• To improve the transparency of deep learning models in MDA prediction.

Main Methods:

• Constructed similarity and meta-path views using similarity and association matrices.
• Employed Graph Convolutional Networks (GCN) for higher-order feature representation.
• Integrated Multi-Channel Attention (MCA) and Semantic Layer Attention (SLA) mechanisms.
• Utilized contrastive learning for dual-graph representation consistency.
• Applied an interpretable Kolmogorov-Arnold Network (KAN) for final prediction.

Main Results:

• iDRKAN significantly outperformed existing computational approaches on two public datasets across multiple performance indicators.
• The method demonstrated a favorable balance between prediction performance and interpretability.
• Case studies confirmed iDRKAN's effectiveness in discovering potential microRNA-disease associations.

Conclusions:

• iDRKAN offers a novel and interpretable approach to microRNA-disease association prediction.
• The dual-graph representation learning and KAN integration provide enhanced accuracy and transparency.
• This method holds promise for advancing biomedical research and clinical applications through improved MDA identification.