Search

We developed models for predicting fish early-life stage (ELS) toxicities oriented to industrial chemicals. The training set was constructed without data from the Office of Pesticide Programs Pesticide Ecotoxicity Database, the main source for the pesticide-biased training set used in our previous work (SAR QSAR Environ. Res. 29:9, 725-742). In addition to the descriptors from the previous study, we also used water solubility to develop the new models, which were evaluated against the test set used in our previous study so that we could focus on the effects of the different training set and the additional descriptor. The statistics for the new models were hardly better than those for the previous models, which suggests, contrary to our expectations, that pesticide-biased data can successfully be used to develop models for predicting the fish ELS toxicities oriented to industrial chemicals. Acute Daphnia magna toxicity was important for the predictive QSAARs in both studies. A distance-based method for defining the applicability domains indicated that water solubility was a key indicator for detecting underestimated chemicals. The comparison of fish ELS toxicities for chemicals presented in different literatures revealed the uncertainty of the experimental data, which may lead to the low predictivity.

Herein, we propose models for predicting fish early-life stage (ELS) toxicity from acute Daphnia magna toxicity and various molecular descriptors. Specifically, eight models were developed with fathead minnow (Pimephales promelas) data and were validated against Japanese medaka (Oryzias latipes) data because the quantity of available Japanese medaka data is much smaller than the quantity of fathead minnow data. The training data set for the models consisted of ELS fathead minnow toxicity data for 77 chemicals; data for 67 of the 77 chemicals originated from the OPP Pesticide Ecotoxicity Database of the US Environmental Protection Agency. The training data were biased toward pesticides. A simple quantitative activity-activity relationship (QAAR) model based on the correlation between fish ELS and acute Daphnia magna toxicities showed good predictivity for the chemicals in the external validation data set relative to the predictivities of the other models in this study. However, goodness-of-fit and robustness were better for quantitative structure-activity-activity relationship (QSAAR) models that included molecular descriptors (such as pesticide-related atoms and substructures as well as molecular weight and three-dimensional-structure-based parameters). A battery approach involving the use of both the QAAR and the QSAARs might enhance the reliability of the estimated values and prevent underestimates.

We evaluated the predictivity and applicability of previously proposed models for the reproductive toxicity of chemicals to Daphnia magna [SAR QSAR Environ. Res. 27:10, 833-850] by using external data from the United States Environmental Protection Agency database ECOTOX. These models were based on quantitative structure-activity-activity relationships (QSAARs) and a quantitative activity-activity relationship (QAAR): the models can be categorized as acute-to-chronic models with (QSAAR) and without (QAAR) structural and physicochemical (e.g. distribution coefficients, log D) descriptors. We found that the QSAAR models were suitable for chemicals with an '-NH 2 attached to aromatic carbon' sub-structure, whereas the QAAR model was better for multicomponent compounds, coordination complexes, tin compounds and straight-chain primary amines. For chemicals with a known specific mode of action (e.g. pesticides and antibacterial agents and their derivatives), toxicity estimation within the acute-to-chronic framework requires special attention. We evaluated the applicability of the models on the basis of the descriptors in the models. We recommend that chemicals be pre-screened before their toxicities are estimated with these models: pre-screening enabled the estimation of the toxicities of some chemicals within the applicability domains of the models.