A Hierarchical Attention-Based Negative Sampling Method for Drug Repositioning Using Neighborhood Interaction Fusion

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Accurate drug-disease association (DDA) prediction is crucial for drug repositioning and developing new therapies.
• Existing DDA prediction methods are limited by insufficient prior knowledge and simplistic negative sampling strategies.
• These limitations hinder the accurate capture of complex drug-disease relationships.

Purpose of the Study:

• To propose a novel model, Hierarchical Attention Mechanism-Based Negative Sampling (HA-NegS), to enhance the prediction of potential DDAs.
• To improve the reliability of negative sample selection in DDA prediction.
• To refine drug and disease representations using graph contrastive learning.

Main Methods:

• HA-NegS computes drug-disease similarity and constructs heterogeneous and homogeneous networks.
• It fuses Graph Convolutional Network (GCN) and Graph Attention Network (GAT) to capture neighborhood features.
• A hierarchical sampling strategy with PageRank and an attention mechanism ensures reliable negative sample selection.
• Graph contrastive learning refines node representations using neighborhood information.

Main Results:

• HA-NegS demonstrated superior performance compared to existing baseline methods on a benchmark dataset for DDA prediction.
• Case studies on Alzheimer's and Parkinson's diseases showcased HA-NegS's potential in identifying novel therapeutic applications.
• The model effectively captures complex drug-disease relationships and refines node representations.

Conclusions:

• HA-NegS offers a significant advancement in predicting drug-disease associations.
• The proposed hierarchical attention mechanism and graph contrastive learning improve prediction accuracy and reliability.
• This approach holds promise for accelerating drug discovery and repositioning efforts.