A QSAR Study For The Acute Aquatic Toxicity Toward Daphnia Magna And Biodegradability Of Some Chemicals

The efficient and convenient prediction of the properties of organics requires replacing,reducing or improving the application of relevant test methods in testing the properties of organics.Quantitative structure-activity/property relationship(QSAR/QSPR)models can relate the biological activity/physical and chemical properties of compounds to the theoretical calculation or experimental description of their chemical structures.So it is necessary to establish QSAR models to predict the relevant properties of organic compounds.This paper mainly includes the following aspects of the research content.(1)Two groups of data were selected in this dissertation.The first group of data is for acute aquatic toxicity data of daphnia magna,which is mainly composed of compounds with carbon,hydrogen and oxygen as main elements,namely ethanol,ketone,phenol and ether compounds.The second group is for the biodegradability data of some organic compounds.The series of the compounds are mainly organic chemicals with benzene ring as the main skeleton,and the substituents are halogen atoms,alkyl groups,phenolic hydroxyl groups,alcoholic hydroxyl groups,ketone groups,and carboxyl groups.The chemicals in the two groups are all non-homologous compounds with complex chemical structures.(2)The molecular descriptors of the two groups of compounds were calculated and selected by ADMEWORKS model builder(version 6.0)software,and then the QSAR models were built with LC50(the concentrations of toxic chemicals causing 50% death of daphnia magna within 48 hours)and BCF(bio-concentration factors)as dependent variables and the descriptors as independent variables.All the variables were normalized prior to generating the models.(3)About 20% of all the normalized samples were randomly selected as external test sets,and then the remaining samples were divided into training sets and internal test sets(4)by using the Sphere-exclusion Algorithms.Finally,the predictions were performed using the k NN weighted regression QSAR models,and their consensus models.In this dissertation,three kinds of k NN weighted regression modeling methods were included,which were different in their calculation formulas for sample weights and end-points.(5)In order to investigate the modeling abilities of the three kinds of QSAR models based on k NN more comprehensively,the composition of the external test sets were changed three times in succession.(6)Based on the thought of the Monte Carlo model population,firstly,multiple data sets were randomly formed to carry out multiple regression modeling on acute aquatic toxicity of daphnia magna and biodegradability of the organic compounds,and then statistical analysis was carried out on the predicted end-points of the obtained external test samples to comprehensively evaluate and compare the advantages and disadvantages of different modeling methods.(7)The three kinds of k NN weighted regression modeling methods,the k NN weighted regression modeling method and the consensus method,the k NN weighted regression modeling method and the regression modeling based on the Monte Carlo model population analysis were compared.The three different k NN weighted regressions were used for one-time modeling.The statistical analysis of the results shows that there is no significant difference in the prediction results among the three kinds of k NN weighted regression models,and the best individual and the consensus models.The consensus modeling method only improved the prediction abilities of some k NN weighted regression QSAR models.The results of Monte model population analysis on multiple modeling of the models show that the above conclusions are more general.Therefore,it is necessary to use Monte Carlo model population for multiple modeling when investigating the prediction abilities of the k NN weighted regression QSAR models.