Prediction of n-octanol-water partition coefficient for polychlorinated biphenyls from theoretical molecular descriptors

Abstract

A quantitative structure-property relationship (QSPR) study was performed to develop models that relate the structures of 133 polychlorinated biphenyls to their n-octanol-water partition coefficients (log K_ow). Molecular descriptors were derived solely from 3D structures of the molecules. The genetic algorithm-partial least squares (GA-PLS) method was applied as a variable selection tool. The partial least square (PLS) method was used to select the best descriptors and the selected descriptors were used as input neurons in neural network model. These descriptors are: Balabane index (J), XY Shadow (SXY), Kier shape index (order 3) (3к), Wiener index (W) and Maximum valency of C atom (VmaxC). The use of descriptors calculated only from molecular structure eliminates the need for experimental determination of properties for use in the correlation and allows for the estimation of log Kow for molecules not yet synthesized. The root mean square errors for ANN predicted partition coefficients of training, test and external validation sets were 0.063, 0.112 and 0.126, respectively, while these values are 0.230, 0.164 and 0.297 for the PLS model, respectively. Comparison between these values and other statistical parameters for these two models revealed the superiority of the ANN over the PLS model.