You are here:
Combinatorial QSAR Modeling of Rat Acute Toxicity by Oral Exposure
Citation:
Zhu, H., T. M. MARTIN, L. Ye, D. M. YOUNG, AND A. Tropsha. Combinatorial QSAR Modeling of Rat Acute Toxicity by Oral Exposure. Presented at Socity of Toxicology Annual Meeting, Baltimore, MD, March 15 - 19, 2009.
Impact/Purpose:
To inform the public
Description:
Quantitative Structure-Activity Relationship (QSAR) toxicity models have become popular tools for identifying potential toxic compounds and prioritizing candidates for animal toxicity tests. However, few QSAR studies have successfully modeled large, diverse mammalian toxicity endpoints (e.g. acute toxicity in vivo). We have applied a combinatorial QSAR approach in the development of robust and predictive models of chemical acute toxicity of rat by oral exposure. To this end, we have compiled a comprehensive dataset of 7,385 compounds with their median lethal dose (LD50) values in the oral exposure test of rats. To compare the predictive power of our models with an available commercial toxicity predictor, we used the same training set of 3,472 compounds as the TOPKAT (Toxicity Prediction by Komputer Assisted Technology) software. The remaining 3,913 compounds in our dataset which were not present in the TOPKAT training set were used as the external validation set for our toxicity models. We have developed 7 different types of QSAR LD50 models for the modeling set. The internal prediction accuracy for the modeling set ranged from 0.52 to 0.96 as measured by the leave-one-out cross-validation correlation coefficient (Q2). The prediction accuracy for the external validation set ranged from 0.24 to 0.70 (linear regression coefficient R2). The use of applicability domain threshold implemented in most models generally improved the external prediction accuracy but at the same time led to the decrease in chemical space coverage. Finally, several consensus models were developed by averaging the predicted LD50 for every compound using all 7 models, with or without taking into account their respective applicability domains. We find that consensus models afford higher prediction accuracy for the external validation dataset with the highest coverage as compared to individual constituent models. Our studies prove the power of collaborative and consensual approach to QSAR model development. The best validated LD50 models developed by our collaboration can be used as reliable computational predictors of in vivo acute toxicity and will be made publicly available from the participating laboratories.
