A Graph Attention Network-based Spatial Decomposition Method for Drug Repositioning | JoVE Visualize

Area of Science:

Computational biology
Pharmacology
Artificial intelligence in medicine

Background:

Computational drug repositioning accelerates drug discovery by identifying new uses for existing drugs.
Graph neural networks (GNNs) show promise for predicting drug-disease associations but face challenges with feature representation and aggregation.
Existing GNN methods often fail to capture higher-order features and inconsistently weight neighbor nodes, limiting prediction accuracy.

Purpose of the Study:

To develop an advanced computational method for drug repositioning using graph neural networks.
To address limitations in current GNN approaches, including feature space conflation and inadequate neighbor node weighting.
To improve the accuracy and efficiency of predicting drug-disease associations.

Main Methods:

Proposed a graph attention network-based spatial decomposition (GATSD) method for drug repositioning.
Initialized drug and disease embeddings in distinct subspaces (drug-similarity, disease-similarity, drug-disease association) using spatial decomposition.
Employed a graph attention mechanism to measure association extents and explore higher-order relationships, incorporating targeted residual connections for personalized feature propagation.

Main Results:

The GATSD method successfully reduced feature space dimensions and initialized embeddings in relevant subspaces.
Graph attention mechanism effectively captured higher-order neighborhood relationships and weighted associations.
Experiments on four benchmark datasets demonstrated that the proposed architecture significantly outperforms current state-of-the-art approaches in drug repositioning prediction accuracy.

Conclusions:

The GATSD method offers a superior approach to computational drug repositioning by effectively learning distinct representations of drugs and diseases.
The spatial decomposition and graph attention mechanisms enhance the ability to predict drug-disease associations accurately.
This work provides a valuable tool for accelerating drug discovery and development.