Quantitative structure–activity relationship modeling of hydroxylated polychlorinated biphenyls as constitutive androstane receptor agonists

Abstract

Hydroxylated polychlorinated biphenyls (OH-PCBs), a series of toxic chemical compounds produced via biotic and abiotic transformation of polychlorinated biphenyls (PCBs), are known to cause endocrine disruption by interacting inappropriately with human nuclear receptors. Due to occurrence of high numbers of inactive OH-PCB congeners recorded in many experimental toxicity studies, it is pertinent to develop rapid and inexpensive QSAR models that can reliably predict the activities of OH-PCB congeners prior to experimental testing. Using a combination of genetic function approximation and multiple linear regression methods, a local QSAR model, consisting of six 2D descriptors (MATS1s, VE3_DzZ, VE1_Dzp, SpMin8_Bhv, SpMax5_Bhi, topoRadius) and two 3D descriptors (RDF95u, RDF45m), was developed from a training set of 44 OH-PCBs. Statistical parameters for fitting ( $${R}^{2}$$ = 0.8902, $${R}_{adj}^{2}$$ = 0.8651, s = 0.2840), cross-validation ( $${Q}_{LOO}^{2}$$ = 0.8201, $${RMSE}_{CV}$$ = 0.3242), and Y-randomization ( $${cR}_{p}^{2}$$ = 0.8019) obtained for the developed QSAR model indicate that the model is reliable, robust, and provides good fit to the data in the training set. The results of external validation carried out on 20 OH-PCBs in the test set also indicate that the developed QSAR model possessed good external predictivity and can be used to predict the agonistic activities of untested OH-PCB congeners to constitutive androstane receptor.