Prediction Of Gas/Particulate Partition Coefficient(K_p) For Polycyclic Aromatic Hydrocarbons And Their Derivatives

Polycyclic aromatic hydrocarbons（PAHs）and their derivatives are a type of volatile persistent organic pollutants（POPs）.After entering the atmosphere,part of it exists in the gaseous form,and the other part is combined with particulate matter aerosols in the air,migrates and transforms with the meteorological movement,and enters other environmental medias,thereby affecting human life and health.The gas/particulate partition coefficient（K_P）of organic compounds is an important parameter to characterize the distribution of organic compounds between the gas phase and the particulate phase,and its value is affected by the n-octanol/air partition coefficient(K_OA).However,both K_P values and K_OA values need to be determined experimentally,which is not only time-consuming and labor-intensive,but also limited by analytical techniques and standard samples of target compounds,making it impossible to determine unknown compounds values.Therefore,it is necessary to develop a quantitative structure-property relationship（QSPR）prediction model to make up for the insufficiency of experiments and provide basic data for the ecological and environmental risk assessment of pollutants.In this study,based on the experimental data of log K_P,using quantum chemical descriptors,multiple linear regression（MLR）and support vector machine（SVM）methods were employed to construct the log K_P prediction models for PAHs and their derivatives.The main research contents and results are as follows:（1）A QSPR prediction model was developed to predict log K_P values of PAHs and their oxygen/nitrogen derivatives.Through the correlation analysis of log K_P values and log K_OA,it was found that they have a strong correlation,with R²=0.801.The average molecular polarizability（α）and the most negative electrostatic potential on the molecular surface(V_s.min)were further selected from nineteen molecular descriptors that characterize information such as molecular electrons and energy using the stepwise MLR method to build model.Simulated external validation and leave-one-out cross validation showed that the model had good fitting ability（R²=0.847）,predictive ability（R²=0.854,Q²=0.847,test set）and robustness(Q²_CV=0.906).The SVM model developed on the basis of MLR model also had good model performance:R²=0.908,Q²=0.853,training set;R²=0.813,Q²=0.818,test set.Mechanistic analysis of the model showed that the main factors affecting the partitioning of PAHs and their oxygen/nitrogen-containing derivatives between gas and particulate phases were intermolecular dispersive interactions and hydrogen bonding.（2）The correlation between log K_P and log K_OA（R²=0.747）of thirty-seven chlorinated/brominated polycyclic aromatic hydrocarbons（Cl/Br-PAHs）was analyzed,and the MLR prediction model was developed using quantum chemical descriptors.There is a significant improvement in performance（R²=0.831）.After simulated external validation and internal validation,the MLR model also had good predictive ability（R²=0.835,Q²=0.809,test set）and robustness(Q²_CV=0.888).The model selectedαand the most negative electrostatic charge of carbon atoms in the molecule（q C^-）,indicating that dispersion and electrostatic interactions dominate the gas/particulate matter partitioning process of Cl/Br-PAHs.On this basis,the SVM nonlinear model was developed,but the application domain Williams diagram diagnoses that the model has an outlier.Reconstructing model after removing the outlier,new SVM model has excellent performance（R²=0.919,Q²=0.925,training set;R²=0.827,Q²=0.830,test set）.The model developed in this study can be used to predict the log K_P values of PAHs and their oxygen/nitrogen,chlorinated/brominated derivatives,and provide basic data for ecological environment safety and health risk assessment of these pollutants.