Drug Repositioning Based on the Enhanced Message Passing and Hypergraph Convolutional Networks.

Area of Science:

• Computational Biology and Bioinformatics
• Drug Discovery and Development
• Network Science

Background:

• Drug repositioning accelerates drug development by identifying new uses for existing drugs.
• Integrating complex biological networks remains a challenge for accurate drug-disease repositioning.
• Existing methods struggle to effectively fuse diverse biological entity data.

Purpose of the Study:

• To propose a novel drug repositioning method, EMPHCN, for enhanced drug-disease association prediction.
• To effectively fuse multi-view drug similarity information and biological network data.
• To improve the accuracy and efficiency of identifying new therapeutic applications for drugs.

Main Methods:

• Constructed homogeneous multi-view information using multiple drug similarity features.
• Employed hypergraph convolutional networks (HGCN) with a channel attention mechanism for intra-domain drug embedding.
• Utilized graph convolutional networks (GCNs) with node and edge embedding (NEEGCN) for inter-domain association extraction, incorporating a heterogeneous drug-protein-disease network.

Main Results:

• EMPHCN achieved superior performance in 10-fold cross-validation, with AUPR reaching 0.593 (T1) and 0.526 (T2), and AUC achieving 0.887 (T1) and 0.961 (T2).
• For novel disease association prediction, EMPHCN demonstrated a 4.3% (T1) and 4.0% (T2) improvement in AUC over the second-best method.
• Case studies confirmed EMPHCN's effectiveness in predicting drug repositioning for breast carcinoma and Parkinson's disease.

Conclusions:

• EMPHCN significantly enhances drug-disease repositioning accuracy by effectively integrating multi-view and inter-domain biological information.
• The proposed method offers a robust framework for discovering new therapeutic indications for existing drugs.
• EMPHCN demonstrates strong potential for accelerating drug discovery and development pipelines.