Predicting the physico-chemical properties of chemicals based on QSPR models
Résumé
Quantitative Structure-Property Relationship models (QSPR) are predictive methods based on correlations between the molecular structures of chemicals and their macroscopic properties. Such methods have been up to now mainly devoted to biological, toxicological applications but their use to predict physico-chemical properties is of growing interest in recent years. In particular, in the framework of the European REACH regulation, the development of such models was recommended as an alternative to experimental tests for reasons. Moreover, such methods represent pertinent tools in screening procedures to select the best performances in any functional properties (e.g. in chemical process) or ensuring at best against hazardous properties (like flammability, explosive or oxidizing properties). To evaluate their reliability, various validation tests are realized. In particular, a robust procedure was proposed by OCDE for their validation for regulatory purpose based on five principles related to the definition of endpoint, the transparency of the model algorithm, its applicability domain, its performances in terms of goodness of fit, robustness and predictive powers, and, if possible, mechanistic interpretation. In that context, INERIS, in collaboration with Chimie ParisTech, develops QSPR models for the prediction of hazardous physico-chemical properties of chemicals like explosive properties of nitro compounds or flammability of amines and organic peroxides. Models were derived according to the OECD validation procedures in view of being submitted to the EU Joint Research Center (JRC) for acceptance or to existing tools (like OECD/ECHA QSAR toolbox for integration.