A Graph Representation Approach Based on Light Gradient Boosting Machine for Predicting Drug-Disease Associations

Area of Science:

• Pharmacogenomics
• Computational Biology
• Bioinformatics

Background:

• Accurate drug-disease association (DDA) prediction is crucial for drug development.
• Current DDA prediction methods often lack diversity in feature extraction techniques.

Purpose of the Study:

• To propose a novel graph representation learning approach for predicting drug-disease associations (DDAs).
• To enhance feature extraction by incorporating network topology and biological knowledge.

Main Methods:

• A heterogeneous network was constructed by introducing protein information.
• Node features were extracted from both network topology and biological knowledge perspectives.
• A light gradient boosting machine (GRLGB) classifier was employed for DDA prediction.

Main Results:

• The GRLGB model demonstrated satisfactory performance on the Bdataset and Fdataset via 10-fold cross-validation.
• Case studies, including anxiety disorders and clozapine, validated the model's reliability.
• The GRLGB approach successfully identified potential novel drug-disease associations.

Conclusions:

• The proposed GRLGB method offers a robust approach for predicting drug-disease associations.
• This novel method addresses limitations in feature extraction homogeneity found in existing DDA prediction strategies.
• GRLGB shows promise in discovering new therapeutic relationships and advancing drug development.