Evaluation of QSAR models for predicting the partition coefficient (logP) of chemicals under the REACH regulation

• Published in: Environmental research 2015-11, Vol.143 (Pt A), p.26-32
• Main Authors: Cappelli, Claudia Ileana, Benfenati, Emilio, Cester, Josep
• Format: Article
• Language: English
• Subjects: 1-Octanol - chemistry; Applicability domain; Chemical Phenomena; Coefficients; Control; Databases, Chemical - legislation & jurisprudence; Errors; European Union; Hazardous Substances - chemistry; Hazardous Substances - toxicity; Health; log Kow; log P; Mathematical models; Mean square values; Models, Chemical; Partitions; QSAR; Quantitative Structure-Activity Relationship; REACH; Roots; Toxicity Tests - methods; Water - chemistry
• ISSN: 0013-9351; 1096-0953
• DOI: 10.1016/j.envres.2015.09.025

The partition coefficient (logP) is a physicochemical parameter widely used in environmental and health sciences and is important in REACH and CLP regulations. In this regulatory context, the number of existing experimental data on logP is negligible compared to the number of chemicals for which it is necessary. There are many models to predict logP and we have selected a number of free programs to examine how they predict the logP of chemicals registered for REACH and to evaluate wheter they can be used in place of experimental data. Some results are good, especially if the information on the applicability domain of the models is considered, with R2 values from 0.7 to 0.8 and root mean square error (RMSE) from 0.8 to 1.5. •Six models assessed to predict logKow of chemicals submitted by industries for REACH.•Best results from VEGA – KOWWIN and VEGA – AlogP on external compounds in AD.•EPI Suite and VEGA – KOWWIN perform better in classification, also considering the AD.•When the AD of the models is considered, accuracy in predictions improves.•The models examined show a general tendency to underestimate the logKow.