In Silico Prediction Of Drug Induced Liver Injury In Traditional Chinese Medicine And Research Of Its Application

ObjectiveOver the last decades,Traditional Chinese Medicine(TCM),regarded as safe and natural,has received more and more attention.Dating back to 2500 years ago,TCM has played an irreplaceable role in Chinese health care system to fight against various diseases and keep health for Chinese people.Despite its long clinical success,recently,an increasing number of hepatotoxicity cases aroused by TCM have been reported,causing widely concern.Thus,there is an urgent need to identify the potential hepatotoxic ingredients in TCM and explore the molecular mechanism of drug induced liver injury(DILI).The aim of this study is to identify potential hepatotoxic ingredients of TCM via building predictive models and explore the molecular mechanism of DILI in TCM by systems pharmacology approach.The research is able to accelerate the process of drug discovery,reduce development costs and also give us an important alert when applying them in clinical practice.MethodsFirstly,all the liver toxic compounds were collected from three public databases,including SIDER,OFFSIDES and Comparative Toxicogenomics Database(CTD).After removing the duplicate compounds,702 liver toxic compounds were finally obtained.Then,the SMILE formats of their structures were downloaded from PubChem.After extracting 100 liver toxic compounds randomly,corresponding decoys were generated in RADER online database with the similarity of 0.75 between liver toxic compounds and decoys.3927 decoys were obtained.Only molecules with molecular weight between 100 and 800 were preserved.A large dataset containing 2476 diverse compounds was constructed,including 619 hepatotoxic compounds and 1857 non-hepatotoxic compounds.All compounds containing hepatotoxic and non-hepatotoxic compounds with the proportion of 1:3,were randomly assigned into training set and test set.Secondly,four types of machine learning methods including artificialneural network(ANN),support vector machine(SVM),random forest(RF)and k-nearest neighbors(kNN)together with four different sets of fingerprints(EState,MACCS,PubChem and SubFP)were utilized to build 16 single classifiers.The performances of all models were measured with the 5-fold cross validation and test set validation.Moreover,to improve the predictive capacity of models,consensus prediction was used to integrate the four best single classifiers after systematic evaluation.Four consensus models named C-1,C-2,C-3 and C-4 were obtained.After evaluating the four consensus models,we found that the performance of C-3 outperformed other consensus models.In addition,the performance between C-3 and four single models was compared and the MCC value of C-3 was higher than any other single models.Then,184 withdrawn drugs were downloaded from DrugBank database to compare the performance between C-3 and Discovery Studio(DS).After browsing the detailed information about the withdrawal causes,we found that 11 of these drugs were withdrawn from the market due to clearly labeled hepatotoxicity.Based on the predicted results of 11 withdrawn drugs from DrugBank,we found that the best consensus model C-3 outperformed the commercial software DS.Subsequently,in order to identify the potential hepatotoxic ingredients in TCM,C-3 was applied to predict the compounds from Traditional Chinese Medicine Systems Pharmacology database(TCMSP).As a result,5666 out of them were predicted as hepatotoxic and the top 10 hepatotoxic herbs including 1042 hepatotoxic compounds were integrated.Furthermore,scaffold analysis was performed via clustering the hepatotoxic ingredients predicted into ten groups.To decipher the hepatotoxicity mechanisms of TCM,systems pharmacology analysis of top 10 herbs and KEGG pathway enrichment of top 20 targets were proposed.Through collecting the compounds and targets as well as constructing different networks containing herb-herb network(H-H network),and compound-target network(C-T network),the hepatotoxicity mechanism of TCM was revealed.Finally,we exemplified hepatotoxicity mechanism of actions by a case study of Chaihu.The network result shows new actions of Chaihu by integrating the hepatotoxic targets and the non-hepatotoxic targets.ResultsConsensus models for drug induced liver injury(DILI)was built to identifypotential hepatotoxic ingredients in TCM and the best consensus model C-3 was found.Then C-3 model was used to screen TCMSP and 5666 hepatotoxic compounds were predicted.The top 10 herbs containing 1042 hepatotoxic compounds was integrated,including Chaihu,Jinyinhua,Yinxingye,Moyao,Chuanxiong,Mahuang,Renshen,Lingzhi,Lajiao and Danshen.The corresponding proportions of the predicted hepatotoxic compounds for each herb to the total compounds of each herb and hepatotoxic compounds in top 10 herbs was calculated.Chaihu has the largest number(141)of hepatotoxic compounds.In addition,scaffold analysis was performed via clustering the hepatotoxic ingredients predicted into ten groups.To explore whether the top 10 herbs discussed above have some similarities in terms of its ingredients,the systems pharmacology analysis of top 10 herbs were proposed.It is found that they do have lots of duplicates in terms of its ingredients.To reveal the most common ingredients in top 10 herbs,frequency analysis of hepatotoxic compounds was performed.Moreover,we chose the top 20 targets(degree?28)in the C-T network to propose the KEGG pathway enrichment.The top 20 targets were enriched in 7 pathways.Among them,5 out of 7 pathways have been experimentally validated to be involved with liver injury.Finally,Chaihu was selected as the case study to decipher hepatotoxicity mechanism of actions and new actions of Chaihu was found.ConclusionIn this study,consensus models were developed on the basis of the four best single classifiers to improve the predictive capability and achieve better performance.C-3 model performed best after systematic evaluation and validation.Subsequently,C-3 model was utilized to screen TCMSP,from which5666 potential hepatotoxic ingredients were predicted.Finally,we integrated the top 10 hepatotoxic herbs and systems pharmacology analysis of hepatotoxic ingredients were proposed to decipher the hepatotoxicity mechanisms of TCM.In summary,this study provides new strategy for DILI and facilitates the discovery and development of new drugs.