Prediction of Drug-Disease Associations Based on Multi-Kernel Deep Learning Method in Heterogeneous Graph Embedding

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Computational drug repositioning accelerates drug development and reduces costs.
• Existing methods struggle to integrate complex biological relationships, limiting drug treatment simulation.
• Identifying novel drug-disease associations is crucial for efficient therapeutic development.

Purpose of the Study:

• To propose HMLKGAT, a heterogeneous graph embedding method for inferring potential drug-disease associations.
• To improve the utilization of complex relationships among biological entities in drug repositioning.
• To enhance the accuracy of predicting novel drugs for diseases.

Main Methods:

• Constructing a heterogeneous information network integrating drug-disease, drug-protein, and disease-protein data.
• Employing a multi-layer graph attention model to capture intricate network associations and derive drug/disease representations.
• Utilizing multi-kernel learning to transform and combine node representations across different feature spaces.

Main Results:

• HMLKGAT significantly outperforms six state-of-the-art methods in drug-related disease prediction.
• Experimental results validate the efficacy of the proposed heterogeneous graph embedding approach.
• Case studies involving five classical drugs demonstrate the practical effectiveness of HMLKGAT.

Conclusions:

• HMLKGAT offers a powerful new approach for computational drug repositioning.
• The method's ability to model complex biological relationships enhances drug discovery pipelines.
• HMLKGAT shows promise for accelerating the identification of effective treatments for various diseases.