Quantitative Structure-activity Relationships Of Paralytic Shellfish Poisoning Toxins

Quantitative Structure-Activity relationships (QSAR) of organic pollutants is of great importance to ecological risk assessment of organic compounds,pollution control and pollution prevention,etc. Quantitative structure-activity/property relationships (QSPR/QSAR) studies have been widely used for prediction of various physicochemical properties and biological activities of organic compounds by different statistical methods and various kinds of molecular descriptors. Thus,QSAR can fill in the data gap of organic pollutants,decrease experimental expenses and especially reduce animal testing. The guidelines for development and validation of QSAR models proposed by the Organization for Economic Co-operation and Development (OECD) were followed. In this study,the performance of variables selection and model optimization in QSAR studies with small sample were compared and discussed,optimal QSAR models were established gradually,and the optimal QSAR models obtained were analyzed and interpreted.In recent years,with the marine environment is worsening,harmful algal blooms occur frequently,resulting in common occurrence of people poisoning and death . In all the red tide toxins , paralytic shellfish poisoning toxins (PSP toxins) have the most widespread distribution,occur the most frequently and have the most harmful effects on human beings. Quantum chemistry descriptors calculated by density functional theory on the B3LYP/6-31G(d) level and modified ant colony algorithm with amended Cp statistical quantity as objective function were used to establish QSAR models for the natural logarithm of semi-lethal doses of the 17 PSP toxins; A model with high correlation coefficients (R2=0.968) and high cross-validation test result (q2=0.858) showed that the model had high precision and good prediction capability. The jackknife method was also used to test the model's stability and reliability. In all descriptors of this work,the eigenvalues of the highest occupied molecular orbital and the next highest occupied molecular orbital play an important role on semi-lethal doses of PSP toxins,indicating that the interaction of molecular orbital and the molecule's reactivity determined the biotoxicity of PSP toxins.The model established in Chapter 2 can't cover all PSP toxins and wasted too much time on the calculation of quantum chemistry descriptors. Problems were solved in Chapter 3. 1751 descriptors and semi-lethal doses of 27 PSP toxins were used to establish QSAR model. In this process,Correlation-based feature selection was used to select features,with leave-one-out cross validation as performance estimator of feature set. 43 descriptors were selected from the 1751 descriptors,with high correlation with target values and low intercorrelation with each other. Principal Component Regression was used to reduce the dimension of the selected feature set,and 10 principal components extracted were used as new features to establish QSAR model. A model with high correlation coefficients (R2=0.891) and high cross-validation test result (q2=0.801) showed that the model had high precision and good prediction capability. Jackknife method was used to test the stability of the model,and 88.9% correlation coefficients falling into between 0.94 and 0.95 showed that the model had strong robustness and reliability. The results indicated that Correlation-based feature selection was fit for selecting features from hundreds and thousands of features,with both incorrelated and redundant features reduced.Leave-one-out cross validation was still used in the model established in Chapter 3. To improve the predictability of the model, independent validation was used in Chapter 4. 1751 descriptors and semi-lethal doses of 27 PSP toxins were used to establish QSAR model. In this process,Correlation-based feature selection was used to select features,with all samples as the training set. 17 descriptors were selected from the 1751 descriptors as the best feature subset. Principal component regression was used in the best feature subset to divide the training set and the test set,and Hotelling T-square was used to remove the outlier. At last,sample 26 was as the outlier,and sample 9,10,12,14,16 and 22 were choosed to constitute the test set. Modified ant colony algorithm and stepwise regression were used respectively to establish QSAR models with the training set,and the results showed that modified ant colony algorithm model was better than stepwise regression model in the fitness of model,residual normal distribution,variables collinearity diagnosis,the robustness and predictability of model. Variable Biowin2,as the most important factor both in the two models,epresents the probability of rapid aerobic biodegradation of an organic chemical in the presence of mixed populations of environmental microorganisms. Carbon with 4 single bonds & no hydrogens,carbamat,fatty alcohol,ester,fatty acid and molecular weight have an indirect effect on pLD50 of the PSP toxins via Biowin2. Topological variables H1v,SIC4 and 5χAv also have important effects on pLD50 of PSP toxins.