You are here:
Using in vitro Dose-Response Profiles to Enhance QSAR Modeling of in vivo Toxicity
Citation:
Sedykh, A., H. Zhu, H. Tang, L. Zhang, A. Tropsha, A. M. RICHARD, AND I. Rusyn. Using in vitro Dose-Response Profiles to Enhance QSAR Modeling of in vivo Toxicity. Presented at Society of Toxicology Annual Meeting, Salt Lake City, UT, March 07 - 11, 2010.
Impact/Purpose:
To develop effective means for rapid toxicity evaluation of environmental chemicals, the Tox21 partnership among the National Toxicology Program (NTP), NIH Chemical Genomics Center, and National Center for Computational Toxicology (NCCT) at the US EPA are conducting a number of quantitative high-throughput screening (qHTS) studies with thousands of chemicals.
Description:
To develop effective means for rapid toxicity evaluation of environmental chemicals, the Tox21 partnership among the National Toxicology Program (NTP), NIH Chemical Genomics Center, and National Center for Computational Toxicology (NCCT) at the US EPA are conducting a number of quantitative high-throughput screening (qHTS) studies with thousands of chemicals. The cell viability qHTS data for an initial set of 1,408 NTP compounds screened in 15-point dose response in 13 cell lines are available in PubChem. We previously showed that biological “descriptors” derived from qHTS IC50 values help to predict in vivo toxicity outcomes of the screened agents using Quantitative Structure-Activity Relationship (QSAR) modeling. However, the full power of the dose-response information of qHTS was not explored. To this end, we have selected 400 qHTS tested compounds, for which binary rodent acute toxicity (i.e., toxic or non-toxic) is known. The classification k Nearest Neighbor (kNN) and Random Forest (RF) QSAR methods were applied using either chemical descriptors alone (conventional models) or in combination with the qHTS-derived biological dose-response profile descriptors (hybrid models). We have also developed special noise-eliminating curve fitting procedures that help address irregularities of the dose-response curves for some compounds and assays. Application of our models to an external dataset resulted in a prediction accuracy of 76% for the conventional models and above 80% for the hybrid models. Moreover, restricting the applicability domain of the hybrid kNN models boosted the prediction accuracy to 86% with only 20% reduction in the chemical space coverage, compared to similar accuracy increase but at the expense of 40% loss in the space coverage for the conventional models. Our study confirms that combining in vitro dose-response profiles with conventional chemical descriptors could considerably improve the prognostic power of QSAR models used for in vivo toxicity prediction. This abstract does not reflect EPA policy.
URLs/Downloads:
Using in vitro Dose-Response Profiles to Enhance QSAR Modeling of in vivo Toxicity (poster) (PDF, NA pp, 192 KB, about PDF)Using in vitro Dose-Response Profiles to Enhance QSAR Modeling of in vivo Toxicity (PDF, NA pp, 8 KB, about PDF)