Quantitative Structure-activity Relationship On Adsorption Coefficient Of Organic Pollutants Between Microplastics And Water Environment

Microplastics have a strong sorption effect on organic pollutants and can act as a transport carrier for organic pollutants,affecting the distribution and transport patterns of organic pollutants in the aquatic environment,promoting the dispersion and bioaccumulation of organic pollutants,and increasing the exposure risk of organic pollutants to aquatic organisms and humans.Therefore,it is of great significance to clarify the sorption capacity of microplastics to organic pollutants in the aqueous environment to assess the environmental and ecological risks and hazards of microplastics and organic pollutants.The adsorption coefficient of organic pollutants between microplastics and water environment(Kd)is often used to characterize the adsorption performance of microplastics.In general,laboratory measurement of Kd may consume a lot of resources in terms of budget and time for testing and evaluation.It is challenging to measure the Kd value of a large number of organic pollutants and new organic pollutants in the environment.Therefore,it is necessary to provide tools that can accurately and quantitatively predict the Kd value of various organic pollutants.In this study,linear and non-linear quantitative structure-activity relationship(QSPR)models were developed using two machine learning algorithms,multiple linear regression(MLR)and multilayer perceptron(MLP),based on temperature and molecular descriptors,to predict the sorption coefficients of organic pollutants between PE microplastics and freshwater(KPE-w)and between PE microplastics and seawater(KPE-sw)at different temperatures.Strictly adhering to OECD guidelines,the models were fully validated internally and externally a nd the application domain of the models was defined.The mechanisms associated with the adsorption of organic pollutants by microplastics were also analyzed.The main study components and conclusions are as follows.(1)A data set consisting of 149 experimental KPE-w values(containing 16 categories of organic pollutants such as benzene,esters,ethers,pesticides,etc.)at 5-30℃ were collected,and MLR and MLP algorithm was used to develop QSPR models for predicting log KPE-w values at different temperatures with six parameters consisting of five molecular descriptors and one temperature as input features obtained by screening.Both models performed well in terms of goodness of fit(Radj2=0.798～0.886),robustness(QLOO2=0.767～0.833,QBOOT2=0.788～0.874)and predictability(Rext2=0.755～0.851,Qext2=0.751～0.844,CCC=0.857～0.916).The performance of the developed MLP-1 model was improved in all three aspects compared to the MLR-1 model.This further confirms the non-linear relationship between the molecular structure information of organic pollutants and the sorption capacity of PE microplastics for organic pollutants.In addition,this study strictly adhered to the OECD guidelines,the applicability domains of these models were defined rigorously.From the mechanistic analysis,it was found that temperature,lipophilicity,ionization potential,molecular size and molecular branching index were the main properties that affect the adsorption capacity of PE microplastics.(2)A dataset covering 47 KPE-sw experimental values(containing 9 types of organic pollutants such as benzene,esters,ethers,pesticides,nitrogen and sulfur compounds,and PAHs)at 18-25℃ was collected and collated.Based on the QSPR model,linear MLR-2 model and nonlinear MLP-2 model were established with five input characteristics(including temperature and four molecular descriptors)as independent variables and KFE-sw experimental values as dependent variables.Internal validation results indicated that both the MLR-2 model and the MLP-2 model were acceptable in terms of goodness of fit and robustness(Radj2=0.871～0.908,QLOO2=0.817-0.771,QBOOT2=0.854-0.896).External validation showed that MLR-2 model and MLP-2 model had strong external prediction ability(Rext2=0.942～0.924,Qext2=0.940～0.917,CCC=0.969～0.959).All data points were within the application domain of the models,and these two models could effectively predict the KPE-sw value of organic pollutants at different temperatures within the application range.Mechanistic explanations suggested that factors such as temperature,lipophilicity,ionization potential,molecular size,branching index and number of branched chains had a significant effect on the sorption capacity of microplastics for organic pollutants.Overall,the developed model can reliably predict the KPE-w and KPE-sw values of new or untested compounds,which can effectively fill in missing experimental values.At the same time,it can provide scientific basis and technical support in the scientific evaluation of the ecological risk of organic pollutants and microplastics and the management of new pollution.