MT-HTI: a novel approach based on metapath2Vec and transformer for herb-target interaction prediction

In the context of the ongoing progress of modern technology, research into Traditional Chinese Medicine (TCM) is being deepened. Advances in modern pharmacology and molecular biology are progressively uncovering the mechanisms of action, efficacy principles, and predictive effects of the components of TCM. Faced with the complexity of TCM components and the intricacies of their mechanisms of action, the traditional compound-target relationship model has limitations in its predictive capabilities. At present, constructing complex heterogeneous graph networks and applying machine learning or deep learning for prediction have become a trend. This paper introduces a novel prediction method based on the efficacy-herb-target-pathway network, with the innovation of incorporating the Metapath2vec. This algorithm trains the model on a heterogeneous graph using manually defined metapaths, capturing the complex relationships within the network more effectively than the traditional node2vec algorithm. In addition, we have developed a custom prediction module based on the transformer architecture, which significantly enhances the accuracy of the predictions. Our method has demonstrated outstanding performance in terms of AUC_ROC, AUC_PR, and F1 evaluation metrics, as evidenced by testing on the collected dataset. This approach not only enhances the accuracy of predictions but also offers a new perspective and tool for predicting TCM targets, thereby adding more practical value to the development of traditional Chinese medicine. MT-HTI is freely available at https://github.comShiLab-GitHub/MT-HTI.