You are here:
Modeling and Predicting Cancer from ToxCast Phase I Data
Citation:
JUDSON, R., D. J. DIX, K. A. HOUCK, T. B. KNUDSEN, I. A. SHAH, D. REIF, M. T. MARTIN, H. M. MORTENSEN, D. ROTROFF, AND R. J. KAVLOCK. Modeling and Predicting Cancer from ToxCast Phase I Data. Presented at Computational Toxicology Board of Scientific Counselors Review, RTP, NC, September 29 - 30, 2009.
Impact/Purpose:
We are using this in vitro data, plus corresponding in vivo data from ToxRefDB, to develop models for prediction and prioritization.
Description:
The ToxCast program is generating a diverse collection of in vitro cell free and cell based HTS data to be used for predictive modeling of in vivo toxicity. We are using this in vitro data, plus corresponding in vivo data from ToxRefDB, to develop models for prediction and prioritization. This poster will focus on a set of machine learning based models that produce toxicity signatures, which are algorithms that yield a toxicity class prediction based on an association between in vitro assay data and an in vivo endpoint, derived from training examples. We demonstrate this approach with a signature for rat liver proliferative lesions using data from the chemicals with rat chronic/cancer data in ToxRefDB, 61 of which are positive for this endpoint. We also demonstrate the use of derived gene and pathway perturbation scores which are more aggregated predictors that can be used in machine learning approaches. Qualitative uses of these perturbation scores are demonstrated with relation to other in vivo endpoints in rodents and humans, including their use in predicting whether a chemical will be a probable human carcinogen. This work was reviewed by EPA and approved for publication but does not necessarily reflect official Agency policy.
URLs/Downloads:
Modeling and Predicting Cancer from ToxCast Phase I Data (poster) (PDF, NA pp, 987 KB, about PDF)Modeling and Predicting Cancer from ToxCast Phase I Data (abs) (PDF, NA pp, 7 KB, about PDF)