Subgraph Neural Networks Enhanced by Global Similarity for Drug Repositioning.

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Drug repositioning accelerates development and reduces costs by finding new uses for existing drugs.
• Graph convolutional networks (GCNs) are increasingly used for drug repositioning, but often fail to capture distinct node roles in graphs.
• Existing GCN methods may struggle with learning effective representations due to overlooking node importance in drug-disease association graphs.

Purpose of the Study:

• To propose a novel subgraph neural network enhanced by global similarity (GSESNN) for improved drug repositioning.
• To address the limitation of existing methods that overlook distinct node roles in drug-disease association graphs.
• To enhance the accuracy of predicting drug-disease associations.

Main Methods:

• GSESNN extracts drug-disease pair subgraphs from a larger graph.
• It employs GCN and sort pooling for subgraph representation learning.
• Global similarity information from GCN is fused with subgraph representations to distinguish similar graph topologies.

Main Results:

• GSESNN outperforms baseline models in drug repositioning tasks.
• The model successfully identified accurate drug-disease associations in case studies for Alzheimer's disease and Gastric Cancer.
• Experimental results demonstrate the model's effectiveness in predicting drug-disease associations.

Conclusions:

• GSESNN offers a promising approach for drug repositioning by effectively learning representations from drug-disease graphs.
• The model's ability to integrate global similarity enhances its predictive power.
• GSESNN shows potential for practical applications in accelerating drug discovery and development.