Coloured chemical image-based models for the prediction of soil sorption of herbicides

Abstract

Herbicides with high soil sorption profiles constitute important organic pollutants leading to detrimental environmental effects, particularly due to prolonged use. Soil sorption is described in terms of log K_OC, the logarithm of the soil/water partition coefficient normalized to organic carbon. This work reports the use of molecular drawings to generate molecular descriptors, which are posteriorly correlated with the log K_OC values of a series of herbicides. These images are two-dimensional projections of chemical structures, with their atom sizes drawn to be proportional to the corresponding van der Waals radii and each chemical element assigned a different colour to distinguish atom types. The progressive changes in the molecular structures explain the variance in the corresponding soil sorption. Unlike previous QSPR studies on soil sorption, the series of herbicides employed in the present study included different classes of compounds (carboxylic acids, ethers, phenols, amines, amides and carbamates) guaranteeing a diverse chemical structural space. The obtained Partial Least Squares (PLS) and Multiple Linear Regression (MLR) based models for the log K_OC values were found to be robust and with high predictive power. Mechanistic interpretation of the effect of different substituents (bonded to the common structural moiety in the herbicides series) on the log K_OC values was performed yielding interesting results. These findings allow greater understanding of the chemical groups (or structural characteristics) responsible for high/low soil sorption, which in turn provides key leads for structural optimization to yield environmentally friendly and equally effective herbicides.