Prediction of hazardous physico-chemical properties of chemicals using QSPR models
Résumé
Quantitative Structure Property Relationships (QSPR) are predictive methods of macroscopic properties based on the molecular structures of chemical substances. Such models were mainly devoted to biological and toxicology applications but are also increasingly used for physic-chemical properties. Within the framework of the European REACH regulation, their development and use are encouraged predictive approaches as an alternative to experimental tests. They are also used within screening strategies to select compounds with target application properties and/or presenting the lower hazardous properties (e.g. flammability, explosibility, oxidizing properties). To estimate the reliability of QSPR models, OECD proposed five validation principles to allow their use within a regulatory framework : definition of the endpoint, unambiguous algorithm, statistical capabilities of the model (goodness-of-fit, robustness, predictive power) and, if possible, an interpretation of the chemical mechanisms. In this context, INERIS develops QSPR models to predict the hazardous physico-chemical properties of chemicals like the explosibility of nitro compounds, the thermal stability of organic peroxides or the flammability of amines. To achieve it, quantum chemical tools are used not only to compute relevant molecular descriptors (notably issued from Conceptual DFT) but also to evidence the underlying chemical mechanisms. These models are developed according to the OECD validation principles toward their acceptance at European level by the JRC (Joint Research Center) for inclusion in their QSAR database and for their integration into existing tools like the QSAR Toolbox of EChA and OECD.