QSAR modeling of different minimum potency levels for in vitro human CAR activation and inhibition and screening of 80,086 REACH and 54,971 U.S. substances

Funding Information: We would like to thank Informatics leader, Senior scientist Ruili Huang from the U.S. National Institute of Health for support on the Tox21 data interpretation. This work was financially supported by the Danish Environmental Protection Agency as well as by the University of California, Los Angeles under the Dissertation Year Fellowship. ; The human Constitutive Androstane Receptor (hCAR) is together with the human Pregnane X Receptor (hPXR) a key regulator of the metabolism and excretion of xenobiotics and endogenous compounds. Inhibition or activation of hCAR by xenobiotics can alter protein expression, leading to decreased or enhanced turnover of both xenobiotics and endogenous substances. Impacts from these alternations can potentially disturb physiological homeostasis and cause adverse effects. Tens-of-thousands of manufactured substances of which humans are potentially exposed are not tested for their potential to inhibit or activate hCAR. In this study, the U.S. Toxicology in the 21st Century (Tox21) high-throughput in vitro assay results for hCAR inhibition and activation were used in a comprehensive in-house process to derive training sets for different potency cut-offs, and to develop suites of quantitative structure–activity relationship (QSAR) models with binary outputs. Final, expanded models, which include substances from the external validation sets, were developed for select minimum potency levels. Rigorous cross- and external validations demonstrated good predictive accuracies for the models. The final models were applied to screen 80,086 EU and 54,971 U.S. substances, and the models predicted around 60% of the substances within their respective applicability domains (AD). Finally, statistical comparisons of hCAR predictions and QSAR predictions for a number of other endpoints related to e.g. Pregnane X, aryl hydrocarbon, estrogen and androgen receptors, as well as genotoxicity, cancer, sensitization and teratogenicity from the Danish (Q)SAR database were made to explore ...