S2: Concordance in Modeling Concentration-Response Data

Concordance in Modeling Concentration-Response Data: Tox21 10k Library Pipeline Comparison

Concordance in Modeling Concentration-Response Data, OpenTox USA 2018

Nisha S. Sipes


National Toxicology Program (NIEHS)


Health Science Evaluator


Nisha S. Sipes, Ruili Huang, Keith R. Shockley, Matthew T Martin, Richard Judson, Keith Houck, Andrew J Shapiro, Joshua Addington, Scott Auerbach, Huixiao Hong, Richard Paules, Jui-Hua Hsieh


Within the field of predictive toxicology, high-throughput data are being generated with potentially different end-user interpretation. One reason for this discrepancy is that the underlying concentration-response models used to analyze the data may be different and hence may yield varying results. Several methods can adequately model dose-response relationships, and the generally-used outcome parameters, such as activity assignments, potency (e.g., AC50 values), and efficacies, can vary. The federal Tox21 collaboration has generated high-throughput screening data on thousands of chemicals and, among the agencies involved, four concentration-response modeling pipelines (3Stage, CurveClass, CurvepwAUC, and TCPL) exist for data analysis. Here, we present results of the evaluation of concordance among the four pipelines. We identify parameters associated with higher concordance between the pipelines and develop a public web application to download and evaluate data taking advantage of the different pipeline analyses.

This abstract does not reflect EPA policy.