On the development of QSPR models for regulatory frameworks: The heat of decomposition of nitroaromatics as a test case

• QSPR models were developed for the heat of decomposition of nitroaromatic compounds.
• Two accurate MLR models were exhibited based on simple constitutional (and topological) descriptors.
• Performances were evaluated by a series of internal and external validations.
• The new QSPR models satisfied all OCDE principles of validation for regulatory use.

Many regulatory frameworks, e.g. related to the Transport of Dangerous Goods, the Registration, Evaluation, Authorisation and restriction of chemicals (REACH) or the Classification, Labelling and Packaging of substances and mixtures (CLP), require the characterization of the hazards of chemicals, which could be complex. In particular, the REACH regulation involves an extensive quantity of works, to gather toxicological, eco-toxicological and physico-chemical properties for a large number of compounds. So, the full characterization by experimental way is time-consuming and cost-expensive. Alternative methods are therefore encouraged to complement experimental tests. The Quantitative Structure–Property Relationships (QSPR) approach is one of the recommended methods, provided that they are developed within the rigorous guidelines proposed by the Organization for Economic Co-operation and Development (OECD). In this context, a series of nitroaromatic compounds has been analyzed to achieve new QSPR models for the prediction of their heat of decomposition respecting the requirements for application in regulatory frameworks.Three multilinear models were obtained upon the set of descriptors considered for their development (constitutional, topological or both) that do not need any preliminary time expensive quantum chemical calculations. They were tested by internal and external validation tests. Good performances for the two ones including constitutional descriptors were obtained in particular in terms of predictive power in a well defined applicability domain (RIN2 = 0.81–0.87). They are easier to apply than our previous quantum chemical based model, since they do not need any preliminary calculations.