Node2Vec-DGI-EL: a hierarchical graph representation learning model for ingredient-disease association prediction

Area of Science:

• Pharmacology and Bioinformatics
• Computational Drug Discovery
• Traditional Chinese Medicine Research

Background:

• Traditional Chinese Medicine (TCM) contains valuable compounds for modern drug development.
• Identifying associations between TCM ingredients and diseases is crucial for drug discovery.
• Existing methods require enhancement for accurate ingredient-disease association prediction.

Purpose of the Study:

• To develop an advanced ingredient-disease association prediction model using hierarchical graph representation learning.
• To leverage complex networks of TCM herbs, ingredients, targets, and diseases for prediction.
• To enhance the prediction of potential therapeutic applications of TCM ingredients.

Main Methods:

• Utilized Node2Vec for initial node embedding extraction in the TCM network.
• Applied Deep Graph Infomax (DGI) to refine node representations and enhance model expressiveness.
• Integrated an ensemble learning method to further improve prediction accuracy.

Main Results:

• The Node2Vec-DGI-EL model achieved superior performance with an AUC of 0.9987 and AUPR of 0.9545.
• Case studies confirmed predictive reliability, identifying potential drug targets for diseases like hypertensive retinopathy and colorectal cancer.
• The model effectively predicts ingredient-disease associations within TCM datasets.

Conclusions:

• The Node2Vec-DGI-EL model demonstrates significant potential for drug repositioning and novel drug development.
• This approach offers a valuable tool for exploring the application value of TCM ingredients.
• The study highlights the importance of graph representation learning in analyzing complex biological networks for pharmaceutical research.