N-octanol-water partition coefficients (log K_OW) of 399 congeners of polychlorinated azoxybenzenes (PCAOBs) determined by QSPR- and ANN-based approach

Abstract

Polychlorinated azoxybenzenes (PCAOBs) theoretically consist of 798 congeners with 399 in cis and 399 in trans configuration. PCAOBs in trans configuration are largerly planar compounds and some are highly toxic and environmentally relevant compared to cis congeners. Trans-PCAOBs can be found as by-side products in 3,4-dichloroaniline and some herbicides. To fulfill gaps in physical and chemical properties of PCAOBs, the values of log K(OW) were determined for 399 congeners of t-CAOB using a computational approach. We used the semi-empirical RM1 in MOPAC and DFT B3LYP in Gaussian 03 methods, artificial neural net (ANN) predictions, and the standardized variables with and without the normal varimax rotation. The models created predicted the values of log K(OW) of all 399 chlorinated derivatives of trans-azoxybenzenes (C-t-AOBs). The values of log K(OW) of C-t-AOBs varied between 5.08 and 5.42 for Mono-, 5.16 and 5.96 for Di-, 5.79 and 6.73 for Tri-, 6.26 and 7.18 for Tetra-, 6.65 and 7.54 for Penta-, 7.13 and 7.94 for Hexa-, 7.20 and 8.20 for Hepta-, 7.96 and 8.32 for Octa-, 8.32 and 8.43 for Nonachloro-t-AOBs and 8.55 and 8.97 for Decachloro-t-AOB. These log K(OW) values were similar per chloro-t-AOB congener and independent of the calculation method. C-t-AOBs have log K(OW) values above 4.5, and what relates to contaminants of low or very low environmentally mobility but a high predilection to the soil and sediment particles and with potential for bioaccumulation. The models that used the standardized variables had smallest errors and higher correlation coefficients compared to the models that based on the normal varimax rotation of standardized structural descriptors. In light of these data, the semi-empirical RM1 calculations in MOPAC software and followed by ANN were a much less time consuming and less expensive compared to the DFT B3LYP method.

A Way towards Reliable Predictive Methods for the Prediction of Physicochemical Properties of Chemicals Using the Group Contribution and other Methods

“…This methodology has been applied to a very wide range of properties, from which we can only name a few. Surfactant cloud points [13], pKa [14], normal boiling points [15] and another one in which the impressive number of 17768 pure chemical compounds were considered and which includes an overview of previous models [16], octanol-water partition coefficients [17] and homogeneous catalysis [18]. The observation though is that accuracy and reliability are often still too limited, something we will further comment on after the next paragraph on the Group Contribution methodology.…”