Estimating the sorption of neutral organic compounds by soil from water using machine learning solute descriptors of the solvation parameter model

When assessing the fate and distribution of chemicals, the soil-water distribution constant is one of the most important physicochemical properties considered in environmental monitoring applications, particularly for emerging environmental contaminants. Since the experimental determination of soil-water distribution constants is time-consuming, tedious, and expensive, various estimation methods are often sought after. The solvation parameter model is a widely used linear free energy relationship model for estimating environmental and biophysical properties. In this work, we processed a large dataset of soil-water distribution constants for 1387 diverse neutral compounds. We estimated machine learning solute descriptors of the solvation parameter model to model the soil-water distribution constant. The proposed method has a prediction error of 0.3 to 0.4 log units in estimating the soil-water distribution constants. We demonstrate that the machine learning solute descriptors for the solvation parameter model can be easily predicted for neutral compounds. The proposed model in this study requires only basic multiplication skills to predict the soil-water distribution constants.