Environmental behavioral controls of polychlorinated biphenyls: prediction of the soil sorption coefficient (Koc) using multiple linear regression

Abstract

The application of the quantitative structure property relationship (QSPR) approach helps to predict physicochemical properties of chemicals from their molecular structures.  They are based on the translating and encoding of input information on theoretical molecular descriptors running by genetic algorithm techniques (GAs). In this research, the focus is mainly on the controlling and understanding the environmental fate and transport mechanisms of polychlorobiphenyls (PCBs) in soils. To achieve this goal, multiple linear regression (MLR) models were handed-down on a series of 188 PCBs to predict their soil sorption coefficient (log K_oc) which is critical to measuring their affinity.  All statistical parameters obtained have indicated that the QSPR-MLR model has a very high predictive ability with a higher coefficient of determination (R² = 0.9984) and lower Root Mean Squared Errors (RMSE = 0.0610).  Moreover, the dominant molecular descriptors are mentioned that the sorption process of these compounds on the surface of soils have been influenced by the size of molecule, polarizability and density.