Sign-Aware Graph Contrastive Learning for Drug Repositioning.

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Drug repositioning accelerates drug discovery by finding new uses for existing drugs.
• Graph Neural Networks (GNNs) are increasingly used for modeling drug-disease associations (DDAs).
• Existing GNN methods often ignore negative links, limiting insights.

Purpose of the Study:

• To propose a novel sign-aware graph contrastive learning approach (SIGDR) for drug repositioning.
• To address challenges in applying sign-aware GNNs to signed biological networks.
• To effectively utilize both positive and negative links in biological networks for DDA identification.

Main Methods:

• SIGDR constructs signed unipartite graphs based on drug and disease similarity.
• A signed bipartite graph is created from annotated DDA data.
• Inter-view contrastive learning enhances node representations using positive and negative subgraphs.

Main Results:

• SIGDR demonstrates effectiveness in identifying drug-disease associations.
• Experiments were conducted on three benchmark datasets.
• The model achieved strong performance under 10-fold cross-validation.

Conclusions:

• SIGDR offers a powerful new approach for drug repositioning using sign-aware graph contrastive learning.
• The method successfully integrates positive and negative links for improved DDA prediction.
• This work advances the application of GNNs in computational drug discovery.