Predicting Potential Drug-Disease Associations Based on Hypergraph Learning with Subgraph Matching.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug repurposing accelerates drug development and disease treatment.
• Predicting drug-disease associations (DDA) is crucial but faces challenges like limited data and noise.
• Emerging technologies like deep learning are increasingly used for DDA prediction.

Purpose of the Study:

• To develop an advanced computational approach for more accurate DDA prediction.
• To address data imbalance and noise issues in existing DDA prediction methods.
• To improve the efficiency of drug discovery and repurposing.

Main Methods:

• Proposed a hypergraph learning with subgraph matching (HGDDA) computational approach.
• Implemented a negative sampling strategy based on similarity networks to reduce data imbalance.
• Utilized a hypergraph Unet module for feature extraction and a hypergraph combination module for DDA prediction via cosine similarity node matching.

Main Results:

• HGDDA demonstrated superior performance compared to existing DDA prediction methods on two standard datasets.
• Validated through 10-fold cross-validation (10-CV), confirming the model's robustness.
• Case studies successfully predicted top drugs for specific diseases, validated against the CTD database.

Conclusions:

• The HGDDA model offers a significant advancement in predicting drug-disease associations.
• The approach effectively handles data imbalance and noise, leading to improved prediction accuracy.
• HGDDA shows strong utility for drug repurposing and accelerating therapeutic development.