You are here:
Using ToxCast in vitro Assays in the Hierarchical Quantitative Structure-Activity Relationship (QSAR) Modeling for Predicting in vivo Toxicity of Chemicals
Citation:
Zhu, H., L. Zhang, A. Sedykh, H. Tang, A. Tropsha, A. M. RICHARD, AND I. Rusyn. Using ToxCast in vitro Assays in the Hierarchical Quantitative Structure-Activity Relationship (QSAR) Modeling for Predicting in vivo Toxicity of Chemicals. Presented at Society of Toxicology Annual Meeting, Salt Lake City, UT, March 07 - 12, 2010.
Impact/Purpose:
Our models could be used to guide the future toxicity studies on the EPA-10K compounds by selecting in vitro assays and prioritizing compounds for in vivo toxicity evaluation. Abstract does not reflect EPA policy.
Description:
The goal of chemical toxicology research is utilizing short term bioassays and/or robust computational methods to predict in vivo toxicity endpoints for chemicals. The ToxCast program established at the US Environmental Protection Agency (EPA) is addressing this goal by using ca. 600 in vitro assays to create bioactivity profiles for a set of 320 compounds with known in vivo toxicity measured in ca. 80 assays. The analysis of this data requires new computational approaches to link chemical structures, in vitro responses and in vivo toxicity effects. We have employed a novel hierarchical QSAR approach to develop predictive models of three ToxCast in vivo multi-generation rat toxicity endpoints: i.e., kidney and liver pathologies, and animal viability index. This approach relies on the relationships between in vitro and in vivo assay results as follows: First, all chemicals are partitioned into two classes based on whether the results of the in vitro and in vivo assays agree (i.e., the compound is found either active or inactive in both types of assays) or disagree (the compound’s annotations in vitro versus in vivo disagree). Second, classification QSAR models for these two classes are developed using Random Forest and Support Vector Machine methods. The resulting QSAR models are used to assign compounds in an external dataset to one of the in vitro/in vivo correlation classes and then predict the associated in vivo toxicity based on the known in vitro response. All the ToxCast bioassays were then ranked based on the external predictivity of the associated models for each in vivo toxicity endpoint. The prediction accuracy for all models was in the range of 61-73% for all three in vivo endpoints, while that achieved by conventional QSAR models was only 50-65% for the same external set.
URLs/Downloads:
Using ToxCast in vitro Assays in the Hierarchical Quantitative Structure-Activity Relationship (QSAR) Modeling for Predicting in vivo Toxicity of Chemicals (abs) (PDF, NA pp, 17 KB, about PDF)Using ToxCast in vitro Assays in the Hierarchical Quantitative Structure-Activity Relationship (QSAR) Modeling for Predicting in vivo Toxicity of Chemicals (poster) (PDF, NA pp, 177 KB, about PDF)