TEMCL: Prediction of Drug-Disease Associations Based on Transformer and Enhanced Multi-View Contrastive Learning

Area of Science:

• Computational biology and bioinformatics
• Drug discovery and development

Background:

• Drug repositioning (DR) is a key strategy for identifying new therapeutic uses for existing drugs.
• Current DR methods often suffer from data sparsity and limited model generalization due to underutilization of biological entities and high-order information.
• There is a need for advanced computational models to capture complex biological data relationships for improved drug-disease association (DDA) prediction.

Purpose of the Study:

• To propose a novel model, TEMCL (Transformer and Enhanced Multi-view Contrastive Learning), for predicting drug-disease associations (DDAs).
• To address data sparsity and enhance model generalization in DR by incorporating high-order biological data features.
• To improve the accuracy and effectiveness of identifying novel DDAs through an advanced computational approach.

Main Methods:

• Utilized a Transformer architecture to extract high-order node features from similarity information.
• Constructed two views: homogeneous hypergraphs and heterogeneous association graphs, incorporating protein nodes and meta-path enhancement to mitigate sparsity.
• Employed Hypergraph Convolutional Network (HGCN) and Heterogeneous Graph Transformer (HGT) for feature extraction, followed by contrastive learning and a Multilayer Perceptron (MLP) for DDA prediction.

Main Results:

• The proposed TEMCL model demonstrated superior performance compared to existing methods in the drug repositioning task.
• Experiments confirmed that TEMCL effectively predicts drug-disease associations, outperforming current state-of-the-art approaches.
• Case studies validated the model's effectiveness, highlighting its potential for identifying novel DDAs.

Conclusions:

• TEMCL offers a powerful and effective computational framework for drug repositioning and DDA prediction.
• The model's ability to capture high-order information and handle data sparsity provides new insights for drug discovery.
• TEMCL represents a significant advancement in identifying novel therapeutic applications for existing drugs.