Development And Application Of A Model For Predicting Gas-Phase Reaction Rate Constants Of Organic Chemicals With Ozone

The gas-phase reaction rate constants of organic chemicals with ozone (A03) are important parameters to characterize the environmental fate and persistence. Since kO3 data are only available for hundreds of organic chemicals, and the experimental determination of organic chemicals is time-consuming and costly, it is necessary to develop an effective predictive model for obtaining kO3 values so as to assess the ecological risk assessment of chemicals. Quantitative structure-activity relationship (QSAR) models have been used for obtaining kO3 as predictive models successfully. Therefore, following the Organization for Economic Co-operation and Development (OECD) guidelines on the development and validation of QSAR models, several QSAR models (the room-temperature (298 K) model and the temperature dependent model) were established for predicting kO3 of organic chemicals. The developed models are of great significance for realizing software-based prediction of environmental parameters. The main results of these models are as follows:(1) Based on quantum chemical descriptors and Dragon descriptors, we developed models to predict kO3 valuesof chemicals at the room temperature (298 K) by employing multiple linear regression (MLR) method and support vector machine (SVM) method. For the MLR model, the adjusted determination coefficient (R2adj), the cross validation coefficient (Q2LOO) and the external validation coefficient (Q2ext) are 0.840,0.809 and 0.800, respectively, indicating that this model has satisfactory goodness-of-fit, robustness and predictive ability. Mechanistic explanation shows that the main factor governing ko3 is the energy of the highest occuptied molecular orbital (EHOMO).kO3 values are also related to the unsaturation, electronegativity and molecular weight of chemicals. In order to describe the nonlinear relationship between kO3 and structural descriptors, the SVM model was established, for which R2adj is 0.938 and Q2ext is 0.826. The Williams Plot was used to describe the applicability domain of two models. Both two models can predict kO3 for a wide range of organic chemicals with these fragments, containing>C=C<,>C=O,-CHO,-COOH,-O-,-OH,-C=O(O)-,-NH2,-NH-,-N-N-,-X(Cl, F) and-S-(2) A MLR model and a SVM model were developed for predicting kO3 values of chemicals at different temperatures, based on total of 399 IogkO3 values for 188 organic chemicals. For the MLR model, R2adj, Q2LOOand Q2cxt are 0.846,0.831 and 0.802 respectively, indicating that this model has satisfactory goodness-of-fit, robustness and predictive ability. Compared with the MLR model, the SVM model represents better predictive performance. Mechanistic explanation shows that the main factors influencing on kO3 values are electron-donating ability, molecular weight, degree of unsaturation, atomic connection information. The applicability domain of two models were visualized by the Williams Plot. Both two models can be used to predict ko3 of many organic chemicals, including alkenes, haloalkenes, oxygen-containing compounds, nitrogen-containing compounds (except primary amines), sulfur-containand compounds and aromatic compounds.We attempt to integrate these models into the "chemical predictive toxicology platform" so as to obtain kO3 values conveniently and effectively. In order to establish an one-click progress from input variables to output results, it is necessary to realize the software-based prediction for rapidly obtaining kO3 values.