Research Of Herb Target Prediction Method Based On Graph Neural Network

Traditional Chinese Medicine(TCM)is a kind of medicine used for prevention,treatment and health care under TCM theory.TCM exerts its efficacy by interacting with targets,which are biological molecules that can be affected by drugs in the body.Correctly identifying the targets of TCM can help researchers further explore mechanisms of TCM.Therefore,predicting targets of TCM effectively is of great significance to clarify the pharmacological mechanism of TCM.In recent years,in order to reduce the time and resource cost of traditional wet experiments,there have been many studies on target prediction of western medicine molecules based on calculation,but TCM can not be directly applicable because of its multi-component and multi-target characteristics.At present,some researchers are based on the idea of herb to component to target for TCM target prediction.However,many herbs have a large number of components,so it is difficult to collect all component data.The others are based on the idea of herb to target,they have developed web-based herb target prediction methods,but failed to make full use of the relationships in herb-target heterogeneous network.Therefore,this thesis proposes TCM target prediction methods based on graph neural network,and implements a TCM target prediction web tool based on the proposed method.The main research work of this thesis is as follows:(1)In view of the problems that previous models do not make full use of potential information of TCM target heterogeneous graph,an approach called HGNA-HTI(Heterogeneous Graph Neural Network with Attention Mechanism for Prediction of Herb Target Interactions)is proposed.HGNA-HTI models TCM targets prediction problem with heterogeneous graph neural network,uses meta relations and attention mechanism to automatically learn the importance of relationship,combines high-order neighbor information through cross layer information transmission,and aggregates semantics into the final feature representations.The overall performance of the HGNA-HTI model on two datasets is better than other benchmarks.Besides,the case study indicates that the HGNA-HTI model also shows better prediction results.(2)Aiming at the problem that the semantic information learned by HGNA-HTI model is relatively implicit and may not be able to give the explanatory judgment basis based on the relevant theories of TCM,a Heterogeneous Graph Neural Network with Metapath Fusion Mechanism for Prediction of Herb Target Interactions called HGMF-HTI is proposed.HGMF-HTI effectively combines the relevant theoretical knowledge of TCM target prediction with metapaths,and gives a clear path reasoning basis for the prediction of TCM targets,and the interpretability is analyzed through case experiments.(3)Based on the research results above,through demand analysis and architecture design,a TCM target prediction web tool based on graph neural network has been designed.The tool provides users with the function of TCM target prediction based on Python application framework Flask.