Prediction of the sorption of organic compounds into soil organic matter from molecular structure

A new model to estimate the soil-water partition coefficient of non-ionic organic compounds normalized to soil organic carbon, Koc, from the two-dimensional molecular structure is presented. Literature data of log Koc for 571 organic chemicals were fitted to 29 parameters with a squared correlation coefficient r2 of 0.852 and a standard error of 0.469 log units. The application domain includes the atom types C, H, N, O, P, S, F, Cl, and Br in various important compound classes. The multilinear model contains the variables molecular weight, bond connectivity, molecular E-state, an indicator for nonpolar and weakly polar compounds, and 24 fragment corrections representing polar groups. The prediction capability is evaluated through an initial two-step development using an 80%:20% split of the data into training and prediction, cross-validation, permutation, and application to three external data sets. The discussion includes separate analyses for subsets of H-bond donors and acceptors as well as for nonpolar and weakly polar compounds. Comparison with existing models including linear solvation energy relationships illustrates the superiority of the new model.