Research On Decision Support System For TCM Pharmacology In Chronic Kidney Disease

The incidence of chronic kidney disease is increasing year by year,which has become a major problem in the medical profession.Chinese medicine prescription has good effects for solving difficult diseases.However,its mechanism is not clear.Network pharmacology studies the treatment principle how the drug interacts the target from the point of molecular level.It thinks that Chinese medicine has multi-ingredients and multi-targets characteristics.Use network pharmacology to research prescription can explain the pharmacological mechanism of prescription and help the doctor make decisions.In this thesis,the mechanism of herbal ingredients is studied to analyze the pharmacological mechanism of prescription by researching how herbal ingredients interact targets to secure diseases at the molecular level,and assist doctors to make decisions.The major three modules in this thesis are interaction relation prediction,network analysis and pharmacological aid decision platform.Use R language to build model,C# language to develop server,echarts plug-in to visualize the networks and graphics.Specific works are as follows:1.The interaction prediction module includes the interaction prediction between ingredients and targets and the interaction prediction between ingredients and diseases.In order to simplify problems,both predictions are treated as the interaction prediction classification problem between ingredients and entities.In this thesis,a fingerprint similarity–based random forest classification model is proposed to identify new molecule-entity relations.In the model,chemical fingerprint is used as the fingerprint feature of the molecule.Fingerprint feature of the entity is calculated using the molecular fingerprint and the interaction relation matrix.Eight methods is used to calculate similarities for a molecule-entity pair as property features for the classifier.The random forest algorithm is used to construct the classification model.Standard,pubchem and maccs fingerprints are used to construct random forest classification model base on enzyme,ion channel,G protein coupled receptor and nuclear receptor datasets,respectively.Five-fold cross validation is used to verify the performances of three different fingerprint models.Average values of AUC(area under curve)for three models reach 99.51%,98.92% and 97.66%,respectively.2.The network analysis module includes the ingredient-target network analysis and the ingredient-disease network analysis.In order to study the mechanism of ingredients to kidney diseases,the ingredient-kidney target network and ingredient–kidney disease network are constructed.In order to study the effects of ingredients to kidney targets and kidney diseases on the network,the pagerank algorithm is used to analysis networks so that important nodes can be identified from nodes with same degrees by different pagerank scores.3.Pharmacological aid decision module helps doctors make decisions base on the points of pharmacology and disease,respectively.In order to help doctors verify the rationality between prescription and syndrome,the correlation between prescription related targets and syndrome related targets is calculated.In order to help doctors analyze and recommend diseases the prescription mainly acts,the ingredient-disease network is analyzed to assist doctor to determine the effectiveness of prescription on the patient.