Predict Endocrine Disrupting Activities Using Ternary Classification Models Based On Nuclear Receptor

Endocrine disrupting chemicals(EDCs)have been attracted extensive attention as they could cause adverse effects with very low dose.With the increasing number and type of EDCs every year,there is an urgent need for healthy risk assessment of EDCs.A major mechanism of endocrine disruption is the action of chemicals as receptor agonists or antagonists through direct interaction with hormone receptors(i.e.androgen receptor(AR),thyroid receptor(TR)and estrogen receptor(ER)).It is feasible to screen and evaluate EDCs using in vivo or in vitro assays,but it becomes prohibitive in terms of both cost and time when dealing with tens of thousands of chemicals.Computational toxicology,especially quantitative structure-activity relationships,provides advantageous alternative methods for identifying EDCs.Thus,a series of researches about assessment of endocrine disrupting activity using computational models were carried out in this study.Polychlorinated biphenyls(PCBs)with 209 congeners are a large family of persistent organic pollutants.Recent studies suggested that PCBs had the potential to cause endocrine disrupting activities.However,PCBs’ thyroid activities have been little studied.We aimed to determine the thyroid-disrupting mechanisms of 209 PCBs through the combination of a novel computational ternary classification model and luciferase reporter gene assay.The reporter gene assay was performed to examine the hormone activities of 22 PCBs via TR and the results revealed that four PCBs exhibited thyroid hormone activity,while eleven PCBs showed anti-thyroid hormone activity.According to experimental data,machine learning methods(i.e.,linear discriminant analysis(LDA)and support vector machines(SVM))were used to build classification models.The optimal model was obtained by using SVM,with an accuracy of 88.24% in the training set,80.0% in the test set and 75.0% in 10-fold cross-validation.The remaining 187 PCB congeners’ hormone activities were predicted using SVM model.The findings revealed that higher chlorinated PCBs tended to have TR-antagonistic activities,whereas lower chlorinated PCBs were agonists.There are increasing empirical data of ER activity,and these data could be applied to develop computational models.In this study,a total of 440 chemicals from the literature were selected to derive and optimize the three-class model,which aimed at classifying the chemicals as agonist,antagonist and no-active via estrogen receptor(ER)through three machine learning methods(linear discriminant analysis,classification and regression trees,and support vector machines).The best model was obtained using support vector machines which recognized agonists and antagonists with accuracies of 76.6% and 75.0%,respectively,on the test set.In addition,109 chemicals from EPA were used to assess their prediction performances by comparing the result of E-screen with the prediction result of classification models.With the similar experimental design and computational principles to ER model,ternary classification models of AR and TR were also developed.Together,we predicted hormone activities of 65 compounds,including 7 pesticides and metabolites using obtained models.This study breaks the fact that previous computational models stop at the binary prediction,and provides more information for the EDCs screening.