Hierarchical Negative Sampling Based Graph Contrastive Learning Approach for Drug-Disease Association Prediction

Area of Science:

• Computational biology
• Bioinformatics
• Network medicine

Background:

• Predicting drug-disease associations (RDAs) is crucial for developing new therapies and repositioning existing drugs.
• Current methods for RDA prediction are limited by reliance on sparse domain knowledge and simplistic negative sampling strategies.
• These limitations hinder the accurate identification of potential drug-disease relationships.

Purpose of the Study:

• To develop an advanced computational model for predicting latent drug-disease associations (RDAs).
• To overcome the limitations of existing methods in handling negative sampling and leveraging complex biological networks.
• To enhance the accuracy and reliability of drug repositioning strategies.

Main Methods:

• A novel hierarchical negative sampling-based graph contrastive model (HSGCLRDA) was developed.
• HSGCLRDA integrates drug, disease, and protein similarities into a heterogeneous drug-disease-protein network.
• The model employs hierarchical structural sampling with PageRank for reliable negative samples and meta-path aggregation for feature extraction using graph convolutional networks (GCNs).

Main Results:

• HSGCLRDA demonstrated superior performance in predicting RDAs compared to existing baseline methods on benchmark datasets.
• The model effectively captures comprehensive interactions between drugs and diseases through global and local feature graphs.
• Case studies highlighted the practical utility of HSGCLRDA in identifying novel potential disease associations for existing drugs.

Conclusions:

• HSGCLRDA offers a significant advancement in predicting drug-disease associations, overcoming key limitations of prior approaches.
• The model's ability to generate reliable negative samples and utilize heterogeneous network information enhances prediction accuracy.
• HSGCLRDA holds practical implications for drug discovery and repositioning by identifying new therapeutic opportunities.