Prediction of soil sorption coefficient of organic pesticides from the atom-type electrotopological state indices.

A group contribution approach based on atom-type electrotopological state indices for predicting the soil sorption coefficient (log KOC) of a diverse set of 201 organic pesticides is presented. Using a training set of 143 compounds, for which the log KOC values were in the range from 0.42 to 5.31, multiple linear regression (MLR) and artificial neural networks were used to build the models. The models were validated using two test sets of 20 and 38 chemicals not included in the training set. The statistics for a linear model with 12 structural parameters were, in test set 1, r2 = 0.79, s = 0.45 and, in test set 2, r2 = 0.74, s = 0.65. These results clearly show that soil sorption coefficients can be accurately and rapidly estimated from easily calculated structural parameters.