You are here:
Consensus Modeling of Oral Rat Acute Toxicity
Citation:
Zhu, H., A. Tropsha, T. M. MARTIN, AND D. M. YOUNG. Consensus Modeling of Oral Rat Acute Toxicity. Presented at QSAR 2010 Workshop, Montreal, QC, CANADA, May 24 - 28, 2010.
Impact/Purpose:
To inform the public.
Description:
An acute toxicity dataset (oral rat LD50) with about 7400 compounds was compiled from the ChemIDplus database. This dataset was divided into a modeling set and a prediction set. The compounds in the prediction set were selected so that they were present in the modeling set used by the commercial toxicity predictor, TOPKAT (Toxicity Prediction by Komputer Assisted Technology). This was done in order to obtain a fair comparison between the predictive power of models generated in this study versus TOPKAT. Five different types of QSAR models (including hierarchical clustering and k-Nearest Neighbor models) were developed for the modeling set. Some of the QSAR models achieved a higher prediction accuracy but at the expense of prediction coverage. Consensus models were developed by averaging the predictions from the other five models. Different coverage levels were achieved depending on how many predictions from the individual models were required to make a prediction. The consensus models achieved the highest prediction accuracy (at a given level of prediction coverage) compared to any of the individual models (including TOPKAT).