| BackgroundAt present,traditional Chinese medicine(TCM)is more and more widely used at home and abroad,and the subsequent adverse reactions/events caused by traditional Chinese medicine also occur from time to time,such as drug-induced kidney injury caused by patients taking TCM such as Mutong and Fangji.The safety of traditional Chinese medicine is an important basis for its clinical application and efficacy.The kidney is an important organ,and the safety of TCM on the kidney has attracted increasing attention and attention.Massive data such as literature database,spontaneous reporting system of the National Adverse Drug Reaction Monitoring Center,and hospital information collection system provide multi-source data for adverse drug reaction mining.Using CiteSpace software to conduct co-occurrence analysis and research on the literature on TCM-induced kidney injury,we can find the TCM and related research directions that have received high attention at present.With the advent of the era of artificial intelligence and big data,deep learning,as a complex machine learning algorithm with autonomous learning and analysis capabilities,is an emerging research method suitable for processing massive data.Using the deep belief network method,based on the data sources reported in the literature,a clinical risk signal early warning model for drug-induced kidney injury caused by TCM can be constructed,and the importance of the explanatory variables in the deep belief network model can be explored by applying the interpretable model SHAP explanatory variables in the deep belief network model.The risk warning signal of sexual kidney injury can provide evidence for the clinical safety of TCM.TCM that lead to drug-induced kidney injury often contain a variety of compounds,and the targets and action pathways of kidney injury are often very complex.The characteristics of biomolecular network research can be based on multiple online database resources.The construction of multi-level networks such as "point-drug potential pathogenic components",and the use of reverse molecular docking technology to carry out the mechanism prediction of TCM-induced kidney injury can provide directions and clues for further research on the mechanism of TCM-induced kidney injury.This study is based on the National Natural Science Foundation of China’s general project "Study on Early Warning Model of Kidney Injury Caused by Post-Marketing Traditional Chinese Medicine Based on Block Chain Fusion and Deep Learning"(No.81973982).ObjectivesBased on the deep belief network method,an early warning model of TCM-induced kidney injury was constructed,and the SHAP interpretable model was used to obtain its risk signal,and the biomolecular network mechanism of drug-induced kidney injury caused by hot TCM was explored,so as to provide evidence for the safe clinical application of TCM.Methods1.In this study,CNKI database(inclusion year:1915),Wanfang database(inclusion year:1900),VIP Information Chinese Science and Technology Periodical Database(inclusion year:1989),Web of Science(inclusion year:1900)4 domestic and foreign databases included literature related to the topic of drug-induced kidney injury caused by TCM from the beginning year to December 31,2021.Obtaining relevant content on drug-induced kidney injury caused by TCM According to the set inclusion and arrangement standards,after reading the bibliography and checking the duplicates,the Chinese and English literatures that meet the theme were finally obtained.Using CiteSpace software to conduct a visual analysis of scientific literature in Chinese and English literature on drug-induced kidney injury caused by TCM Co-occurrence views and salient words of co-citations,co-occurrence views,clustering views and salient words of keywords were analyzed for the research institutions,authors,countries,and literature co-citations of TCM-induced kidney injury.Finally,get the research hotspots of TCM and research frontier issues.2.This study searched CNKI database,Wanfang database,VIP information Chinese science and technology journal database and related medical records of drug-induced kidney injury and non-drug-induced kidney injury caused by the use of hot traditional Chinese medicines and their compound preparations in the treatment of diseases.The data from the medical records of drug-induced kidney injury caused and not caused by the use of hot traditional Chinese medicines and their compound preparations to treat diseases were extracted.Missing values were handled using mean substitution and dummy variables.Build a deep belief network(DBN)early warning model.Applying the interpretable model SHAP explains the importance of variables in a deep belief network model.Obtain the risk signal of drug-induced kidney injury caused by TCM3.In this study,TCMSP and TCMID databases were retrieved to extract the active ingredients of research hotspots of TCM Relevant literatures from CNKI database,Wanfang database,VIP information Chinese science and technology journal database,Pubmed,and Web of Science databases were searched to supplement the potential pathogenic components of TCM The potential targets of active ingredients were predicted using Swiss Target Prediction and SEA database.The STRING database was used to establish an analysis network of target-protein interactions related to drug-induced kidney injury caused by hot traditional Chinese medicines.And use Cytoscape v3.6.1 software to display it visually.The network was topologically analyzed using the NetworkAnalysis plug-in,and the mean degrees of freedom of each target were obtained.Targets with more degrees of freedom than the mean degree of freedom were selected as core targets.Based on the webgestalt website,GO annotation and KEGG+Reactome pathway enrichment analysis were performed on the obtained hot TCM-induced kidney injury targets.Autodock Vina 1.5.6 software was used to conduct molecular docking between the key targets of hot TCM-induced kidney injury and the obtained active ingredients to verify the binding activity of Guanmutong active ingredients and key targets.Results1.A total of 327,805 articles were obtained,including 323,928 Chinese documents and 3,877 English documents,and 2,458 Chinese documents and 272 English documents were obtained through manual analysis of the bibliography.There were 248 Chinese literatures and 248 English literatures in line with the theme of drug-induced kidney injury caused by TCM The area with the most publications in Chinese literature using CiteSpace software was Beijing,the author with the most publications was Chen Yipu,the institution with the most publications was Beijing University of Traditional Chinese Medicine,and the research hot spot of TCM was "Mutong"("Guanmutong"accounts for 88.9%).Drug-induced kidney injury includes acute kidney injury,interstitial nephritis,acute and chronic renal failure and other diseases.Research hotspots such as clinical application,risk factors,pathway analysis,safe medication,risk control,pathogenic mechanism,and pathogenic components(2017～2022);the countries,authors and institutions with the largest number of publications in English literature were China,KAISHUN BI,Shenyang Pharmaceutical University.And the top five keywords were "aristolochic acid nephropathy","rat","traditional chinese medicine","chinese herb nephropathy","dna adduct".The research results show that domestic and foreign scholars were paying more attention to the mechanism of drug-induced kidney injury caused by TCM of the genus Aristolochia containing aristolochic acid.2.The clinical literature reports using Guanmutong were retrieved,and 22305 documents were retrieved.After checking the duplicates and reading the full text,866 documents were finally screened.Finally,83 medical records of Guanmutong caused drug-induced kidney injury and 1627 cases of drug-induced kidney injury caused by Guanmutong were extracted.Injury case data.The data were preprocessed by means of mean substitution and dummy variables to fill in the missing data.31 variables were used as the modeling variables of the DBN model.The learning rate of the model was set to 0.1.80%of the data set after data processing was divided into training set,while 20%was divided into test set.And the number of layers was set to 3 layers.The layer was set with 64 nodes,and the DBN model was obtained.The accuracy of the training set was 100%.The accuracy of the test set was 99.96%.The average cross-validation score of ten-fold cross-validation was 99.54%.The precision was 99.70%,and the recall rate was 100%.The F1 score was 0.9985.And the AUC value(area under the ROC curve)was 1.00.The interpretable SHAP model showed that gender,vomiting,nausea,age,and fatigue were important risk signals for Guanmutong-induced kidney injury.3.The study obtained 22 potential active ingredients such as aristolactam i,aristolochic acid,aristolochic acid ii,etc..It predicted 1146 pathogenic targets of Guanmutong and 859 targets of drug-induced kidney injury.By taking the intersection of the two types of targets,147 targets of Guanmutong-induced kidney injury were obtained.34 core targets were obtained by topological analysis of the network using the NetworkAnalysis plug-in and screening by the MCODE plug-in.Using the webgestalt website to conduct GO and KEGG+Reactome target pathway analysis on the core targets,20 GO biological processes including the positive regulation of cell movement were enriched,and 20 KEGG+Reactome key pathways including the immune system were obtained.The molecular docking results of Guanmutong were compared with the molecular docking results of the positive control drug neomycin,which can cause kidney damage.10 key targets of IL2,MMP9,MTOR,PTGS2,TLR4,TNF and VEGFA may be the key components of Guanmutong-induced kidney injury.Conclusions1.According to the quantitative analysis of scientific literature,"Guanmutong" was a research hot spot of traditional Chinese medicine for TCM-induced kidney injury,the hot spot method of clinical research was early warning model,and the frontier method of mechanism research was network pharmacology and molecular docking.2.The constructed TCM(Guanmutong)-induced kidney injury DBN early warning model AUC value=1.0,indicating that the model evaluation efficiency was good,which can explain the model SHAP shows that gender,nausea,vomiting,age,fatigue were important risk signals of Guanmutong-induced kidney injury.3.Network pharmacology and molecular docking technology found that ferulic acid and daucosterol may be the key components of Guanmutong-induced kidney injury.Guanmutong may lead to drug-induced kidney injury through the immune system,Th17 cell differentiation pathway,hepatitis B,pertussis,toxoplasmosis,leishmaniasis and other related biological processes/pathways.. |
