In Silico Prediction Of Anticancer ROCK Ⅱ Inhibitors And Evaluation Of Cardiovascular Toxicity For Anticancer Agents

ObjectiveThe development of anticancer drugs is a long process with high risk and large investment.During this process,the discovery of lead compounds and the evaluation of drug toxicity are the two key aspects.In this paper,in silico approaches including machine learning,combined classifier and molecular docking,were applied to advance these two aspects in the discovery of anti cancer drug.We first constructed a virtual prediction model of inhibitor targeting ROCK II kinase,which is a potential drug target for cancer treatment.Effective mechanism of ROCK Ⅱ inhibitors was also discussed.Second,a combined model was constructed for the most common cardiovascular toxicities in antitumor drugs and the cardiovascular toxicity of various antitumor drugs was evaluated.Based on the results of the study,13,139 ingredients covering 499 Chinese herbs in Chinese Pharmacopoeia(version 2010)were virtual screening for their ROCK Ⅱ inhibitory activity,and the positive ones were further evaluated for their cardiovascular toxicity.MethodsExperiment 1:Construction of anticaner ROCK II inhibitor prediction model and its mechanism st.udyFirstly,983 compounds with inhibitory activity against Rho-related protein kinase Ⅱ(ROCK Ⅱ)were collected from the BindingDB database and related literatures.The descriptors were calculated using MOE 2010 and PaDEL-Descriptor software and optimized by F-score and Linear Forward Selection(LFS)method.After that,16 preliminary prediction models were constructed by using four kinds of machine learning classification algorithms(knearest neighbors,random forest,naive Bayes,and support vector machine).Then,three kinds of fingerprints were added to improve the performance of the best four models.The prediction capability of the models were assessed by 5-fold cross validation,test set validation,and external test set validation,and the best models were applied to predict the ROCK II inhibitory activity of ingredients from 499 Chinese herbs.The binding mode between ROCK II inhibitor and receptor was discussed by privileged substructure analysis and molecular docking,while main interactions were revealed by comparing the docking interaction of t.he most potent and the weakest ROCK II inhibit.ors t.o revealed its pharmacodynamic mechanism.Finally,the bingding ability between ROCK II and ingredients from Chinsee herbs were further evaluated.Experiment 2:Construction of combined classifiers for identify drug induced cardiovascular complications and its application on anticancer agentsFirstly,information on drug-induced cardiovascular complications covering the five most common cardiovascular toxicities including hypertension,heart block,arrhythmia,cardiac failure and myocardial infarction was collected from four authoritative databases(CTD,SIDER,Offsides and MetaADEDB).After that,combined classifiers were constructed by integrating five machine learning algorithms(logistic regression,random forest,k nearest neighbors,support vector machine and neural network).The classifiers were applied to the evaluation of cardiovascular safety of anticancer agents,including 63 clinical anticaner agents and potential ROCKⅡ inhibitors from Chinese herbs predicted by experiment 1.The evaluation results were verified by clinical research data,human pluripotent stem cell-derived cardiomyocyte assay data and literature experimental data.ResultsExperiment 1:Construction of anticaner ROCK II inhibitor prediction model and its mechanism studyAfter comprehensive evaluation and comparison,MFK + MACCS and MLR + SubFP which are based on the combination of optimized descriptors and molecular fingerprints have the most prominent predictive ability.Both of their MCC value reached 0.925 on external test set validation,which accuracy(Q)reach 0.972.Their sensitivity(MFK + MACCS is 0.910,MLR + SubFP is 0.933)and specificity(MFK + MACCS is 0.993,MLR + SubFP is 0.985)also performed well.Considering their outstanding and well-balanced performance,MFK + MACCS and MLR + SubFP are regarded as the best two ROCK Ⅱ inhibitors prediction models.They were applied for screening 13,139 ingredients and 240 of them were predicted as inhibitors.The advantage fragment analysis and molecular docking technology revealed that the hydrophobic interaction as well as hydrogen bonding,especially between Glul70 group and amine group,may be the main important driving forces for the activity of ROCK II inhibitors.Based on the results of molecular docking results and mechanism research,the binding ability between 3 ingredients and ROCK II were further evaluated,and ingredient M0L008822(thaliphylline)exhibited high binding level with the protein active site.Experiment 2:Construction of combined classifiers for identify drug induced cardiovascular complications and its application on anticancer agentsIn this experiment,a total of 180 single prediction classifiers of cardiovascular toxicity were constructed,and five combined classifiers were generated by integrating the four best single classifiers based on neural network for each cardiovascular adverse event.Although single classifiers performed well on 5-fold cross-validation with AUC values ranging from 0.647 to 0.809,combined classifier performed even better,with AUC values ranging from 0.784 to 0.842.The best four single classifiers and combined classifiers were applied to evaluated drug-induced cardiovascular complications of 63 clinical anticancer agents.The prediction results were validated using the experimental data from clinical studies,human pluripotent stem cell-derived cardiomyocyte assays,and literatures.The successful rate reached 87%.The combined classifiers successfully revealed serval anticancer agent-induced cardiovascular complications which have not yet been listed on their FDA profiles.Anticancer agents targeting Bcr-Abt,DNA topoisomerase and microtubule system also had been alerted higher risk of cardiovascular toxicity.In addition,240 potential ROCK II inhibitors from Chinese herbs predicted by experiment 1 were further evaluated for their cardiotoxicity,and 14 of them were considered as low cardiac safety risk ones.This study provides guidance and early warning on drug development and clinical application.ConclusionIn this paper,in silico approaches were applied for the two important aspects of the development of anticancer drugs,including the discovery of novel anticancer compounds(ROCK II inhibitors)and the evaluation of cardiovascular complications for anticancer agents.In experiment 1,the fisrt reported ROCK Ⅱ inhibitors predictive models utilizing machine learning approaches were constructed,and their outstanding performance were confirmed by multiple evaluation methods.Moreover,privileged substructure analysis and molecular docking revealed the effect mechanism and important functional groups of ROCK Ⅱ inhibitor.In experiment 2,an in silico framwork based on combined classifiers with high accuracy was proposed for the evaluation of drug-induced cardiovascular complications.This method was applied to the cardiovascular toxicity prediction for anticancer agents.Combined classifiers successfully discovered several novel anticancer agent-induced cardiovascular complications and uncovered diverse mechanisms that may have high risk of cardiovascular toxicity.Based on the results of the two experiments,the inhibitory activity of ROCK Ⅱ protein and the cardiotoxicity of the 13,139 ingredients from Chinese herbs were predicted,which discovered ingredients with high development value.In general,this paper provides new strategies for discovering lead compounds and evaluating toxicity during the process of anticancer drug development especially for Chinese medicine,and will also provide some enlightenment to the discussion of therapy mechanism and early warning of cardiovascular toxicity for clinical anti-tumor treatment.