The aqueous solubility of some herbicidal by-side toxic impurities: predicted data of the 399 chlorinated trans-azoxybenzene congeners.

The quantitative structure - property relationship (QSPR) and the artificial neural networks (ANNs) methods were used to estimate aqueous solubility (log S and μg/L) of polychlorinated trans-azoxybenzenes (PCt-ABs). These QSPR and ANN models are based on geometry optimalization and quantum-chemical structural descriptors, which were computed on the level of density functional theory (DFT) using B3LYP functional and 6-311++G** basis set in Gaussian 03 software and the semi-empirical quantum chemistry method for property parameterization (RM1) in the molecular orbital package (MOPAC) software. The predicted solubility of PCt-AOBs by RM1 and DFT models and depending on a congener varied within a homologue class between 47-19498 and 371-1738 μg/L for Mono-; 33-11481 and 7.9-3630 μg/L for Di-; 6.1-4786 and 4.7-12882 μg/L for Tri-; 1.3-1174 and 0.3-14791 μg/L for Tetra-; 0.4-646 and 0.1-38904 μg/L for Penta-; 0.1-155 and 0.2-63096 μg/L for Hexa-; 0.2-27 and 0.1-646 μg/L for Hepta-; < 0.1-6.2 and 0.8-282 μg/L for Octa-; 0.6-2.6 and 0.8-12 μg/L for NonaCt-AOBs; and 1.2 and 0.5 μg/L for DecaCt-AOB, respectively. Both computational models used were characterized by good predictive abilities and small errors, while calculations by RM1 method were highly competitive compared to a much more time-consuming and expensive method by DFT.