Node2Vec-DGI-EL: a hierarchical graph representation learning model for ingredient–disease association prediction
Leifeng Zhang, Xin Dong, Shuaibing Jia, Jianhua Zhang · Bioinformatics Advances · 2025
Motivation: Traditional Chinese medicine, as an essential component of traditional medicine, contains active ingredients that serve as a crucial source for modern drug development. To explore the potential application value of traditional Chinese medicine ingredients, this study utilizes the complex network formed between herbs, ingredients, targets, and diseases, and proposes an ingredient-disease association prediction model (Node2Vec-DGI-EL) based on hierarchical graph representation learning. The model first utilized Node2Vec to extract node embedding vectors, serving as the initial features for the network nodes. Then, DGI was applied to further refine the node representations, enhancing the model's expressive power. Finally, an ensemble learning method was integrated to further improve prediction performance. Results: The proposed model significantly outperformed existing methods, achieving an AUC of 0.9987 and an AUPR of 0.9545. Case studies further validated the reliability of the model's predictive results. Specifically, triptonide exhibited a binding energy of -9.62 kcal/mol with PGR, a core target of hypertensive retinopathy, while methyl ursolate showed a binding energy of -9.71 kcal/mol with NFE2L2, a core target of colorectal cancer. The Node2Vec-DGI-EL model focuses on traditional Chinese medicine datasets, effectively predicting ingredient-disease associations. It demonstrates significant application value and can assist in drug repositioning and novel drug development. Availability and implementation: The code and data are available at https://github.com/wayfarer569/Node2Vec-DGI-EL.