Modeling of retention time for polychlorinated biphenyl congeners in human adipose tissue using quantitative structure–retention relationship methodology

Abstract

In this study, quantitative structure–retention relationship methodology was employed for prediction retention time of the 209 polychlorinated biphenyl (PCBs) in human adipose tissue on the DB-1701 column. The molecular descriptor was calculated, and most important ones were selected by stepwise multiple linear regression (SW-MLR) approaches. These descriptors are symbolically represented as BEHp2, JGI1, MATS3e, RDF050m which are an independent variable used for generation of linear and nonlinear model by multiple linear regression (MLR) and support vector machine (SVM) methods, respectively. Comparison of statistical parameters of these models indicated that the SVM model performed better than MLR which represents the nonlinear relationship between the structural feature and retention time of PCBs on this chromatographic condition. The relative standard error (SE) of the training and test set for the SVM model was 0.67 and 0.76, and the square correlation coefficients (R 2) were 0.977 and 0.970, respectively. The mean effect analysis indicated that the polarity and structural features of studied PCBs are important factors responsible for chromatographic retention of these chemicals at the same chromatographic condition.