You are here:
Quantitative Predictive Models for Systemic Toxicity (SOT)
Citation:
Truong, L., G. Ouedraogo, S. Loisel-Joubert, H. Nocairi, AND M. T. Martin. Quantitative Predictive Models for Systemic Toxicity (SOT). Presented at SOT annual meeting, San Diego, CA, March 22 - 26, 2015. https://doi.org/10.23645/epacomptox.5178820
Impact/Purpose:
poster presented at SOT annual meeting in San Diego, CA on March 24, 2015.
Description:
Models to identify systemic and specific target organ toxicity were developed to help transition the field of toxicology towards computational models. By leveraging multiple data sources to incorporate read-across and machine learning approaches, a quantitative model of systemic toxicity was developed. The data sources included high-throughput screening (HTS), in vivo toxicity, and reverse toxicokinetic (RtK) data. The initial quantitative model of systemic toxicity used chemical fingerprints and read across to predict lowest effect levels (LEL) for 578 chemicals. Additionally, three independent random forest models were developed using chemical descriptors, biologically-grouped HTS data, and RtK-converted HTS data. A consensus approach was applied to the 4 models, with 5-fold cross-validation, repeated 10 times, resulting in a RMSE of 1.07, and a R2 of 0.35. Furthermore, qualitative models classifying liver and kidney toxicity were developed for 535 chemicals using k-nearest neighbor and random forest approaches on chemical fingerprints and HTS data. Approximately 58% of the chemicals were identified as liver toxicants and 34% as kidney toxicants for modeling purposes. Consensus models were developed for each data source and resulted in a positive predictive value between 0.7-0.8 and balance accuracy of 0.6-0.8, depending on the machine learning method and target organ. Herein, we have identified consensus models that predicted systemic LEL or classified chemicals as liver and/or kidney toxicants utilizing the power of chemical, in vitro and in vivo data. This abstract does not necessarily represent EPA policy.
