GADRC: a graph-based approach for drug repositioning with deep residual networks and computational feature-guided...

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repositioning (DR) accelerates therapeutic discovery by identifying new uses for existing drugs.
• Current computational DR methods face limitations in capturing complex drug-disease relationships and efficiently using negative samples.
• Shallow network architectures and the vanishing gradient problem hinder the modeling of multi-level interactions.

Purpose of the Study:

• To develop an advanced computational method, GADRC, to overcome limitations in current drug repositioning strategies.
• To enhance the extraction of both local and global features for drugs and diseases.
• To improve the utilization of negative samples in drug-disease association prediction.

Main Methods:

• Employs a synergistic architecture of graph convolutional networks and graph attention networks for feature extraction.
• Introduces a deep residual network with identity connections to address network depth degradation.
• Utilizes a feature-guided undersampling strategy and weighted cross-entropy loss for efficient negative sample mining.

Main Results:

• GADRC consistently outperforms existing methods on three benchmark datasets for drug repositioning tasks.
• The method successfully captures local drug structural features and global disease pathway features.
• Experimental results demonstrate improved utilization efficiency of negative samples.

Conclusions:

• GADRC offers a robust and effective approach for identifying novel drug-disease associations.
• The method's ability to learn multi-level interactions and handle negative samples significantly advances computational DR.
• Case studies on Alzheimer's disease and breast cancer highlight GADRC's potential for real-world therapeutic discovery.