Prediction of partition coefficients for complex environmental contaminants: Validation of COSMOtherm, ABSOLV, and SPARC



Prediction of partition coefficients is essential for screening of environmentally relevant compounds. Prediction methods using only the molecular structure as input are especially useful for this purpose. In the present study, the authors validated 3 prediction method—COSMOtherm, ABSOLV, and SPARC—which are based on more mechanistic approaches than most other quantitative structure–activity relationships. Validation was based on a consistent experimental data set of up to 270 compounds, mostly pesticides and flame retardants. The validation systems included 3 gas chromatographic (GC) columns and 4 liquid/liquid systems that represent all relevant types of intermolecular interactions. Results revealed that the overall prediction accuracy of COSMOtherm and ABSOLV is comparable, whereas SPARC performance is substantially lower than the other methods. For instance, the root mean squared error for the 4 liquid/liquid partition coefficients was 0.65 log units to 0.93 log units for COSMOtherm, 0.64 log units to 0.95 log units for ABSOLV, and 1.43 to 2.85 log units for SPARC. In addition, version and parameterization influences of COSMOtherm on the prediction accuracy were determined. Environ Toxicol Chem 2014;33:1537–1543. © 2014 SETAC