QSAR Study On Adsorption Behavior Of Pharmaceutically Active Compounds By Sludge In Wastewater

Widespread use of drugs lead to the continuous accumulation of pharmaceutical active compounds（Ph ACs）in the water environment,which eventually presents"pseudo-persistent"pollution and endangers the environment.The main source of pharmaceutical active substances in the environment is sewage treatment plants.The removal of pharmaceutical active ingredients in sewage treatment plants is mainly through biochemical degradation and sludge adsorption,of which sludge adsorption is the most important way.The solid-liquid partition coefficient(_（9）is a key index to measure the adsorption of drugs by sludge,which is of great significance to evaluate the extent of drug adsorption by sludge.However,the value measured by experiments is inefficient and costly.It is difficult to meet the needs of ecological risk assessment and water pollution control only by using traditional experimental measurement methods to obtain the₍（9）value of drugs.Therefore,it is necessary to develop a prediction model for sludge adsorption of drug active substances₍（9）.The development of quantitative structure-activity relationship（QSAR）prediction model has become an important way to solve such problems.In this study,the₍（9） value of drug active substances adsorbed by sludge was collected through experimental measurements or reported literature data.According to the OECD guidelines,details of model development were all considered.The molecular structure descriptors were calculated using Dragon software.Based on different modeling methods,such as multiple linear regression（MLR）,support vector machine（SVM）and projection pursuit regression（PPR）,the predicted QSAR model was constructed.The main contents and conclusions are as follows:（1）The QSAR model of drug active substances adsorption by primary sedimentation tank sludge was constructed.A data set containing 87 kinds of pharmaceutical active substances in primary sedimentation tank sludge and their corresponding₍（9）values were collected from the published papers,and the QSAR model was established based on MLR algorithm,SVM algorithm and PPR algorithm,respectively.The training set and test set were divided by random splitting and principal component analysis（PCA）,respectively.The results show that in the proportion of 4:1,the sample distribution of training set and test set divided by PCA grouping method is relatively uniform,and the model results are also better than the results of random splitting,indicating that PCA method is more suitable for dividing training set and test set.The optimal model is composed of six molecular descriptors.The application domain of the optimal model is characterized,and it is found that there is an abnormal point in the optimal model.After removing the outliers,use the same molecular structure descriptor to re-establish the MLR model.The statistical parameters of the new model are as follows:the prediction result for the test set is R²=0.792,the root mean square error RMSE=0.390,and the absolute average relative error AARD=13.481%,respectively.Using SVM algorithm to build model,the prediction results are more accurate than that of the linear model,which are R²=0.854,RMSE=0.396,AARD=12.398%,respectively.Compared with the PPR algorithm,the PPR model improved the R² of the test set to 0.862,while RMSE and AARD decreased to 0.369 and 12.697%,respectively.The mechanism analysis shows that the partition coefficient is related to many factors such as the molecular structure,hydrophilicity,polarity and functional group of drug active substances,among which the polar surface area plays an important role.（2）A QSAR model of activated sludge adsorption of drug active substances was constructed.The data set contains 108 active substances of activated sludge drugs.Based on the eight molecular descriptors selected by multiple linear stepwise regression,the QSAR model was established using MLR algorithm,SVM algorithm and PPR algorithm,respectively.The results show that the PPR model is more accurate than MLR model and SVM model in terms of prediction ability.For the test set,PPR gives R²=0.800,RMSE=0.497 and AARD=21.650%,respectively,which are better than the MLR model’s R²=0.735,RMSE=0.558,AARD=27.388%,and also better than the SVM model’s R²=0.754,RMSE=0.545,AARD=21.255%.The Kd value of active substances absorbed by activated sludge is mainly related to many factors such as molecular hydrophobicity,electron donor capacity,functional groups and the number of R-CX-R fragments in molecular structure,etc.In this study,a variety of algorithms were used to establish QSAR models,and the corresponding statistical parameters indicate that the QSAR models established in this study exhibit good stability not only in terms of goodness of fit but also in terms of prediction ability.The mechanism analysis can provide a theoretical basis and basis for the behavior of sludge adsorption of drug active substances.