Abstract:
Polyhalogenated compounds (PHCs) such as polybrominated and polychlorinated diphenyl ethers (PBDEs/PCDEs) are an important chemical group because of their environmental persistence, high hydrophobicity, and bioaccumulation in humans. Their physico-chemical properties such as n-octanol/air partition coefficient (log Koa) and n-octanol/water partition coefficient (log Kow) and toxicities are of fundamental importance to gain a better understanding of the environmental fate and behavior of these compounds. In this study, several Quantitative Structure-Activity Relationship/Quantitative Structure-Property Relationship (QSAR/QSPR) models were developed on log Koa of PBDEs, log Kow of PBDEs and PCDEs, and the aryl hydrocarbon receptor relative binding affinity (log RBA) of PBDEs by employing Heuristic Method (HM) and Multiple Linear Regression (MLR). Descriptors used were from DRAGON 5.4, SPARTAN 06, and CODESSA 2.2 software and the Characteristic Root Index (CRI) program. All the best models were internally validated for their performance using the leave-one-out procedure and scrambling of the responses. External validation was provided by splitting the data sets into training and test sets either using random division in terms of property modeled or Kohonen network considering the size of the data sets. Of the models developed log Koa and log Kow models were validated externally by using test sets. log RBA model could not be validated externally because of a lack of RBA data. EHOMO and Eaq from SPARTAN, and the CRI appeared to be significant descriptors for the developed log Kow models of PBDEs/PCDEs. The CRI also appeared to be an important parameter in modeling log Koa of PBDEs. The statistical quality of all the models for polyhalogenated diphenyl ethers is compared to those of the previously published models using the same experimental data and found to be superior to those models. All the QSAR/QSPR models were developed taking into account the OECD principles for validation, for regulatory purposes, of QSAR. This implied internal and external validations, the analysis of the applicability domain (AD) and, when possible, a mechanistic interpretation of the models.
