QSAR Models For Aqueous Reaction Rate Constants Of Organic Pollutants With Hydroxyl Radical And Singlet Oxygen

Reactions with hydroxyl radical(·OH)and singlet oxygen(1O2)are important pathways governing transformation and degradation of organic pollutants in the aquatic environment.Second reaction rate constants of ·OH and 1O2 reactions(kOH and k1O2)are significant parameters for environmental persistence evaluation and ecological risk assessment.To acquire aqueous kOH and k1O2 values,experimental determination is inefficient and costly,which fails to satisfy the demands of ecological risk assessment for organic pollutants with diverse classes and forms,while developing quantitative structure-activity relationship(QSAR)models is an important approach.In this study,databases for aqueous kOH and k1O2 values of organic pollutants were constructed based on experimentally determined or reported data.QSAR models,employing quantum chemical as well as Dragon molecular structural descriptors and multiple algorithms[such as multiple linear regression(MLR),partial least squares(PLS)and support vector machine(SVM)],were then established.Main contents and results are as follows:(1)QSAR models for predicting aqueous kOH were established.A database covering aqueous kOH values of 526 organic pollutants was constructed,and an MLR model was then established.The model exhibits good goodness-of-fit[the value of determination coefficient(R2)is 0.805],robustness[the values of explained variance for cross-validation and bootstrapping method(Q2CV and Q2BOOT)are 0.797 and 0.791,respectively]and predictive ability[the values of external determination coefficient(R2ext)and explained variance(Q2ext)are 0.802 and 0.801,respectively].For the pollutants in the model,their aqueous kOH values are mainly affected by their electron-donating ability,also related to other factors such as electronegativity,polarity and functional groups.For the 4 classes of pollutants(saturated alkanes,alkenes and alkynes,phenols,saturated alcohols)with specific ·OH reaction mechanisms,their ·OH reactivity is affected by factors such as the number and electron-donating ability of the ·OH reactive sites.(2)A QSAR model for predicting aqueous kOH values of antibiotics at different dissociation forms was established.kOH values of 9 fluoroquinolones(FQs)and 11 sulfonamides(SAs)under 3 pH conditions and at 3 dissociation forms were determined via competition kinetics experiments.Results indicate that kOH values of the 20 antibiotics vary significantly among different pH conditions and dissociation forms.An MLR model for predicting kOH values was established,with good goodness-of-fit(R2=0.805),robustness(Q2CV=0.778,Q2BOOT=0.770)and predictive ability(R2ext=0.820,Q2ext=0.816).Model analysis indicates that aqueous kOH values of the antibiotics are mainly affected by the number of the ·OH reactive sites(piperazine rings,quinoline rings of the FQs,and benzene rings of the SAs),charge distribution and molecular polarity.(3)QSAR models for aqueous k1O2 were established.A database covering aqueous k1O2 values of 226 organic pollutants was constructed,and QSAR models were then established.For the models,stepwise MLR and genetic algorithm(GA)were employed to select descriptors,while MLR,PLS and SVM algorithms were employed to construct models.All of the models possess good goodness-of-fit(the values of R2 ranging from 0.708 to 0.966),robustness(the values of Q2CV ranging from 0.671 to 0.858)and predictive ability(the values of R2ext and Q2ext ranging from 0.721 to 0.891).Descriptors obtained from GA lead to better model performance than those from stepwise MLR.Significantly,the non-linear SVM models perform better than the linear MLR and PLS models.For the pollutants in the models,their aqueous k1O2 values are mainly affected by their electron-donating ability,also related to other factors such as polarizability,electronegativity and functional groups.For the 4 classes of pollutants(pollutants containing carbon-carbon double bonds,phenols,amines and sulfur-containing pollutants)with specific 1O2 reaction mechanisms,non-linear regressions employing the energy of the highest occupied molecular orbital exhibit higher fitting ability than linear regressions.In summary,the current study combined simulated experiments with QSAR models to explore in silico methods for aqueous kOH and k1O2,which contributes to filling data gaps in ·OH and 1O2 reaction kinetic parameters of ionizable pollutants.Futhermore,this study provides fundamental data and tools for evaluating and predicting environmental persistence and ecological risk of organic pollutants.