Herb-phenotype association prediction based on graph neural network

Abstract

Natural products represented by herbs have played a crucial role in the discovery of drugs for complex diseases. However, herbs contain a multitude of active compounds, making it challenging to consider their individual effects and interactions. Graph neural networks can be employed as a solution to address these limitations. By enabling information exchange among nodes in a graph, graph neural networks can incorporate connectivity relationships and obtain updated information. In this study, we represented herbs as networks of their constituent compounds and applied a graph neural network model to predict natural product-phenotype relationships. The results demonstrated that our approach outperformed traditional machine learning models in terms of predictive performance. Furthermore, a detailed analysis of the top 10 herbs associated with diabetes in a case study provided evidence for the effectiveness of the model designed in this study. This not only highlights the potential of graph neural networks for representing the diverse compounds within herbs but also validates their ability to accurately predict herb-phenotype associations.