|
چکیده
|
In this work, the liquid chromatographic retention times of some organic pollutants were modeled and predicted by the quantitative structure retention relationship (QSRR) approach. The data set consists of the retention times of 36 organic pollutants. The genetic algorithm-partial least square (GA-PLS) was used as a featured selection technique; the artificial neural network (ANN) and the support vector machine (SVM) were used for generation of the QSRR models. Descriptors which were selected by GA-PLS are mean atomic Van der Waals volume, molecular weight, number of double bonds, number of acceptor atom for H-bond, and topographic electronic descriptor. These descriptors were used as inputs for developed ANN and SVM models. After generation and optimization of ANN and SVM models, the models were used to calculate the retention time for internal test set. The root mean square errors of the GA-ANN model were 0.89 and 1.22 and the root mean squares of the GA-SVR model were 3.08 and 1.67 for training and test sets, respectively. Also, for the further evaluation of the credibility of the models, the leave-seven-out cross validation test was done. The statistical parameters of these tests were Q2 ¼ 0.905 and SPRESS ¼ 7.5 for the GA-ANN model and Q2 ¼ 0.690 and SPRESS ¼ 6.55 for the GA-SVR model. These results reveal the suitability of ANN in the prediction of liquid chromatographic retention times of organic pollutants
|