Prediction of retention times of pesticide residues in cultivated lands using linear and nonlinear chemometric methods : Stepwise Multiple Linear Regression and Artifical Neural Network

QSPR approach was done to provide the relationship between the structures and of the retention times of 208 pesticide residues in the cultivatid lands. The molecular structure of the compounds were drawn by ChemDraw 8.0 software, was optimized by HyperChem 7.0 software, after that the molecular discriptors of the molecules were calculated by Dragon 2.1 software. Then the number of variables was decresed by stepwise multiple linear regression method. The data set based on the principal of normal distribution, were divided into two series training and test sets. Multiple linear regression create models with six descriptor which are molar refractivity and spanning tree number and 3D-MoRSE descriptors and 3D Petitjean shape index and 2nd component accessibility directional WHIM index / unweighted and 3rd component accessibility directional WHIM index / weighted by I-state. Statistical values of F, R2Pred, R2Train and R2Test obtained for models using SW-MLR were 106, 0.905, 0.902, and 0.921 respectively. The obtained model was evaluated using the technique of cross-validation leave-one-out (LOO) method. Then, to increase the efficiency of the model, non-linear method artificial neural networks (ANN), was used. Predicted value of retention times and statistical parameters obtained from the two methods were compared. The results showed that the artificial neural network (ANN) is superior to the other methods. R2pred values for the method of artificial neural network was 0.960 respectively.

کلیدواژه ها:

Quantitative structure property relationship ، Retention time ، Pesticides ، Cultivated land ، Multiple linear regression ، Artificial neural network