Estimating soil partition coefficient of organic chemicals using M5 tree decision model

The organic carbon partition coefficient, Koc, represent the extent to which an organic substance partitions between the soil/sediment and aqueous phases, therefore it is useful to estimate the transport of chemicals in aquatic and soil systems. In this study a dataset consisting of 530 organic chemicals were gathered from literature and databases. The purpose of this study was to develop linear models correlate the octanol/water partition coefficient, Kow, and water solubility, Sw, with Koc. The M5P Machine learn method as it is implemented in the WEKA system, was used to derive 2 equations that correlated log Koc with log Kow, log Sw. All models have two leafs that separate lipophilic chemicals from low hydrophobic chemicals and were evaluated with correlation coefficient (r) and root mean squared error (RMSE). Howbeit the correlation coefficient and the errors of two models are close but the best model developed in this paper is the linear model derived by using logKow with r = 0.9035 and RMSE = 0.641.

Keywords:

soil partition coefficient , M5 tree decision model , Hydrophobicity , water solubility , Structure–activity relationships