Prediction of Acute Herbicide Toxicity in Rats from Quantitative Structure-Activity Relationship Modeling.

Extensive use of herbicides raises concerns about adverse effects on the environment. Studies on toxicity of herbicides are few and relatively old. The purpose of this study was to use multiple linear regressions (MLR) and Multilayered Perceptron artificial neural networks (MLP-ANN) to predict the oral acute toxicity (half-maximal lethal dose [LD50]) of a diverse set of 62 herbicides on rats. Quantitative structure-activity relationship (QSAR) models obtained by using relevant descriptors showed good predictability. Primary contributions to toxicity were the following descriptors: HATS0m, HATSe, and nS. Comparison of results obtained using the MLP-ANN model with those from the MLR model revealed the superiority of the MLP-ANN model. Statistics for prediction of oral acute toxicity for MLR and MLP-ANN were, respectively: R2=0.855, RMSE=0.270; and R2=0.960, RMSE=0.118. Comparison of validation results with those of other studies have shown the superiority of the model developed in this work. [ABSTRACT FROM AUTHOR]