A Hypergraph Convolutional Network With Explicit High-Order Interaction Information Extraction for Drug...

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Drug repositioning accelerates drug development by finding new uses for existing drugs, reducing costs and risks.
• Hypergraph Neural Networks (HGNNs) are increasingly used for drug repositioning due to their ability to model complex relationships.
• Existing HGNN methods often fail to capture diverse relations and high-order interactions adequately.

Purpose of the Study:

• To propose HGCNDR, a hypergraph convolutional network designed for explicit high-order interaction extraction in drug repositioning.
• To enhance the modeling of diverse relations and high-order interactions among drugs and diseases.
• To improve the accuracy and efficiency of identifying novel drug-disease associations.

Main Methods:

• Developed HGCNDR, incorporating a relation-aware hypergraph convolution and a Hadamard product strategy for high-order interactions.
• Constructed feature graphs and a hypergraph using drug/disease similarity and association networks.
• Employed Graph Convolutional Networks (GCNs) for feature graph embeddings and specialized operations for hypergraph embeddings.
• Introduced a consistency constraint to preserve semantic commonalities between embeddings.

Main Results:

• HGCNDR demonstrated competitive performance against several established baseline methods.
• Experimental results indicate superior ability in retrieving actual drug-disease associations.
• Case studies on Alzheimer's disease and Breast carcinoma validated the method's predictive power.

Conclusions:

• HGCNDR effectively addresses limitations in existing HGNN methods for drug repositioning by explicitly modeling high-order interactions.
• The proposed model shows significant promise for identifying novel therapeutic applications of existing drugs.
• HGCNDR offers a valuable tool for accelerating drug discovery and development pipelines.