Evaluation of hazardous properties of substances by QSPR models : case of thermal stability of organic peroxides

Due to the fast development and availability of computers, predictive approaches are increasingly used in the evaluation process of hazardous substances complementary to experimental approaches or in a first step of screening of these hazardous properties. Their use was in particular recommended as alternative to experimental testing by the REACH regulation to complete the lack of knowledge on properties for existing substances that have to be registered before 2018 (upon quantities). Among the proposed predictive approaches, Quantitative Structure Property Relationships (QSPR) are powerful methods that allow for the prediction of macroscopic properties from the knowledge of the only molecular structure of substances. In this context, INERIS developed QSPR models for the prediction of hazardous physico-chemical properties (explosibility, thermal stability or flammability) of chemical substances. In particular, works devoted to the development of QSPR models to predict stability properties of industrial organic peroxides were performed within the French PREDIMOL (molecular modeling prediction of physicochemical properties of products, 2010-2014) project [1] funded by National Research Agency and within a new European project, entitled HAZPRED (Predictive methods for determining the decomposition properties of hazardous substances: from development to experimental verification, 2015-2018) financed by the SAF€RA consortium dedicated to industrial safety [2]. After establishment of an inventory of existing experimental data from literature, additional experiments were performed to develop robust databases. Then, existing models applicable to these compounds were identified and tested. Finally, new QSPR models based on MLR and genetic algorithms were developed leading to the first reliable models for the thermal stability of organic peroxides [3] validated according to the OECD principles for their use into regulatory context. Simple models (including only constitutional and topological descriptors) could be proposed for integration into the ECHA/OECD QSAR Toolbox [4], as the one proposed for explosibility of nitro compounds [5] that was implemented in the 3.3 version of the Toolbox in December 2014 representing up to now the only included model for hazardous physico-chemical properties. Works are currently performed to predict the SADT (self-accelerating decomposition temperature) of organic peroxides. These methods also allow identifying hazards of organic peroxides as early as possible in their development