MVSGDR: multi-view stacked graph convolutional network for drug repositioning.

Area of Science:

• Computational Biology
• Pharmacology
• Network Science

Background:

• Drug repositioning (DR) is a cost-effective drug development strategy.
• Current computational DR methods struggle to integrate local substructure patterns with global network semantics.
• Existing methods often rely on data augmentation to address information gaps in drug-disease associations (DDAs).

Purpose of the Study:

• To present a novel DR framework, multi-view stacked graph convolutional network (MVSGDR), to overcome limitations in current computational DR approaches.
• To improve the prediction accuracy of drug-disease associations (DDAs).

Main Methods:

• Developed MVSGDR, a novel DR framework incorporating three innovations.
• Multi-view stacked module for depth-wise feature enhancement via hierarchical aggregation of multi-hop neighborhood interactions.
• Bi-level subgraph transformer module for breadth-wise analysis of DDAs using METIS-partitioned subgraphs.
• Negative sampling balancing strategy to mitigate sample imbalance through synthetic negative samples.

Main Results:

• MVSGDR demonstrated superior performance in extensive 10-fold cross-validation experiments across four benchmark datasets.
• Statistically significant improvements were observed compared to existing DR methods.
• Case studies successfully identified previously unreported DDAs with supporting literature evidence.

Conclusions:

• MVSGDR offers a powerful new framework for drug repositioning.
• The proposed methods effectively synergize localized substructure patterns with global network semantics.
• MVSGDR shows significant potential for identifying novel therapeutic applications of existing drugs.