Research And Implementation Of Drug-Target Relationship Model Of Traditional Chinese Medicine Prescriptions Based On Graph Neural Network

Traditional Chinese medicine is an important part of the traditional culture and has a history of thousands of years.Chinese medicine treatment methods include external treatment methods(acupuncture,moxibustion,massage)and internal treatment methods(traditional Chinese medicine,diet therapy).Traditional Chinese medicine treatment is a kind of internal treatment method,which is the main and most commonly used treatment method of traditional Chinese medicine treatment.Traditional Chinese medicine treatment is based on prescriptions.Doctors determine the patient’s syndrome after syndrome differentiation,determine the treatment method according to the syndrome,and then combine the treatment method according to the nature,flavor and meridian distribution of the traditional Chinese medicine.Follow the principle of monarch,minister,assistant and envoy to formulate prescriptions,adjust the body’s energy metabolism distribution to achieve therapeutic effect.Traditional Chinese medicine acts on corresponding disease targets through active ingredients,and achieves the purpose of medication by regulating the physiological metabolic process in the body through pathways.Most traditional Chinese medicines are natural medicines,and their active ingredients and targets are complex and difficult to understand.Existing network research methods cannot accurately capture the complex characteristics of heterogeneous networks.The thesis takes the drug-target relationship model of traditional Chinese medicine as the research object,carries out the research on the target prediction of traditional Chinese medicine,the discovery of active ingredients of traditional Chinese medicine and the prediction of prescription indications,and designs and realizes the analysis platform of the target point relationship of traditional Chinese medicine prescription.The main tasks of the research are as follows:1.Aiming at the problems that the target prediction of traditional Chinese medicine depends on the difficult components and cannot distinguish heterogeneous nodes,the Herb Target predict based on Meta-Path Networks(HTMNet)based on meta-path embedding captures the characteristics of traditional Chinese medicine and The connection between western medicines,learning representative neighborhood node features through different meta-paths.The test results show that HTMNet has achieved good performance on the ACM data set that compares the network embedding performance and the collected traditional Chinese medicine target data set.It has achieved Macro-F1 and Micro-F1 of 0.8871 and 0.8874 respectively on ACM,while in The AUROC on the collected Chinese medicine target data set is 0.8984.2.Aiming at the problems that the existing analysis methods of active ingredients of traditional Chinese medicine only use complex network technology and require manual combing,the Herb Ingredient predict with Co-contrastive learning(HIC)analysis model based on collaborative comparative optimization is proposed.The network embeds the original image separately,maps two different types of embeddings to the same feature space,and learns features adaptively and collaboratively.The test results show that HIC achieves 0.8904 and 0.8871 for Macro-F1 and Micro-F1 respectively on the Freebase data set,and it is verified on the constructed Chinese medicine ingredient data.The results show that HICNet can predict the active ingredients of Chinese medicine,and its AUROC and AUPR accuracy rates are respectively Reach 0.8784 and 0.8642.3.Aiming at the problems of ignoring targets and difficulty in processing complex network information existing in traditional Chinese medicine symptom methods,the prescription indication prediction method based on multi-layer graph convolution(MultiLayer based Formula Symptom Predict,MLFSP)was proposed.Information extracts the homogeneous node relationship characteristics of symptom pairs and Chinese medicine pairs network,and extracts the heterogeneous node association characteristics between Chinese medicine symptoms through the message passing mechanism,and integrates the two to predict prescription indications.The test results show that the Macro-F 1 of MLFSP on the Cora,Citeseer,and PubMed datasets are 0.8317,0.6985,and 0.8605.4.Based on SpringBoot,Vue.js,Element and other frameworks,using the BS architecture model,using Java language and MySQL database to design and realize the relationship analysis platform of Chinese medicine prescriptions and medicine targets to assist the modernization research of Chinese medicine,the platform has system management,data management,Chinese medicine target Functions such as point prediction,analysis of active ingredients of traditional Chinese medicine,prediction of prescription indications.