QSPR studies on the aqueous solubility of PCDD/Fs by using artificial neural network combined with stepwise regression

A practicable quantitative structure property relationship (QSPR) model for predicting aqueous solubility, Sw, of 23 polychlorinated dibenzo-p-dioxins and polychlorinated dibenzofurans (PCDD/Fs) was developed. Linear artificial neural network (L-ANN) was used to develop the calibration model of Sw. The input variables of L-ANN were selected from 11 structural descriptors of the investigated PCDD/Fs by using stepwise regression. Leave one out cross validation and split-sample validation were carried out to assess the predictive performance of the developed model. The results of leave one out cross validation and split-sample validation are both satisfactory, which verify the reliability and practicability of the developed model. It is demonstrated that L-ANN combined with stepwise regression is a practicable method for developing QSPR model for Sw of PCDD/Fs. Additionally, stepwise regression is shown to be a practicable approach for the selection of input variables when developing a QSPR model with L-ANN.