The partitioning tendency of pesticides, in these study herbicides in particular, into different environmental compartments depends mainly of the physic-chemical properties of the pesticides itself. Aqueous solubility (S) indicates the tendency of a pesticide to be removed from soil by runoff or irrigation and to reach surface water. The experimental procedure determining aqueous solubility of pesticides is very expensive and difficult. QSPR methods are often used to estimate the aqueous solubility of herbicides. The artificial neural network (ANN) and support vector machine (SVM) methods, every time associated with genetic algorithm (GA) selection of the most important variable, were used to develop QSPR models to predict the aqueous solubility of a series 80 herbicides. The values of log S of the studied compounds were well correlated with de descriptors. Considering the pertinent descriptors, a Pearson Correlation Squared (R2) coefficient of 0.8 was obtained for the ANN model with a structure of 5-3-1 and 0.8 was obtained for the SVM model using the RBF function for the optimal parameters values: C = 11.12; ? = 0.1111 and ? = 0.222.

Pesticide use in agriculture can cause undesirable effects on humans and the natural environment. Physicochemical properties of pesticides play an important role in determining its distribution and fate in the environment. Chemometric methods can be used to describe how the physicochemical properties vary according to the characteristics of the molecular structure expressed in terms of appropriate molecular descriptors. Quantitative Structure-Property Relationship (QSPR) models can also provide a general overview of the molecular structure that influences these properties. Henry's law constant (H) is an important property for predicting the solubility and vapor-liquid equilibrium of pesticides. Genetic algorithm/ multi-linear hybrid approach was used to model the log H of 48 pesticides belonging to four chemical classes: ureas, triazines, carbamates and aryloxyalkanoic acids. The 5 explanatory variables model selected is robust and has good fitness and good predictive ability. Two influential points which reinforce the model and an outlier were highlighted. The model can be used to predict the Henry's law constant of pesticides falling in the applicability domain of our model.

No abstract