Biochemistry
Shahin Ahmadi; Shahram Lotfi; leila shamakhi; Ali Azimi
Abstract
The increasing use of pesticides following the rising in agricultural, demand has threatened non-target organisms such as avian species and disrupted the ecological system. Therefore, considering the application methods and the nature of chemical pesticides, testing their toxicity level on birds, and ...
Read More
The increasing use of pesticides following the rising in agricultural, demand has threatened non-target organisms such as avian species and disrupted the ecological system. Therefore, considering the application methods and the nature of chemical pesticides, testing their toxicity level on birds, and protecting various species of endangered birds is an essential requirement from the point of view of ecosystem safety. In this study, quantitative structure-toxicity relationships modeling was done for the first time to estimate the toxicity of 244 types of pesticides on five different species of birds consist of bobwhite quail (C. virginianus), mallard duck (A. platyrhynchos), house sparrow (P. domesticus), ring-necked pheasant (P. colchicus), and Japanese quail (C. japonica). All data were randomly divided into four series including active training, passive training, calibration, and test sets. Hybrid optimal descriptors, resulting from the combination of quasi-SMILES descriptors and hydrogen- suppressed graph (HSG) based on a new target function, were used to generate QSTR models. Four target functions (TF0, TF1, TF2, TF3) were used to develop QSTR models and the predictive potential of these models was evaluated using a validation set. The QSTR models designed using TF3 target function with the range of R2 = 0.7218-0.8131 and Q2 = 0.7031-0.7878 for the validation set were statistically the best models. Statistically, the best model is model number six, with R2 values for active training, passive training, calibration, and validation sets equal to 0.836, 0.852, 0.806, and 0.813, respectively. The mean absolute error (MAE) values for the sets of active training, passive training, calibration and validation are 0.371, 0.342, 0.409 and 0.362, respectively, indicating the accuracy of the model created to predict the toxicity of pesticides against five species of endangered birds. From the results of this modeling, important descriptors were identified for increasing and decreasing the average effective toxicity concentration (pLD50) of pesticides. Using the QSTR models obtained from this study, it becomes possible to predict the toxicity (pLD50) of new pesticides even before their synthesis by only having the SMILES symbol of the pesticides, which can help to reduce time, resources, costs and the need for laboratory animals.
