Predicting neurological targets for chemical neurotoxins using ToxCast in vitro data and read-across within QSAR Toolbox

Abstract

Transportation fuels consist of a large number of species that belong to different families of compounds. Surrogate fuel representations have been formulated to better understand their fundamental chemical composition and to emulate combustion properties. These descriptions are formulated using experiments or through computations, which has thus led to the existence of two different notions of surrogates. There is further distinction of concepts through the use of physical and chemical surrogates, which are designed to emulate those specific properties. Although several surrogate design methodologies have been proposed in literature, they do not incorporate information on experimental uncertainty. By addressing this issue, it is shown that this information is crucial for the reliable construction of surrogates through computations. To incorporate physical fuel properties, a consistent approach through the use of the recent ASTM D2887 distillation curve standard is discussed. Then, a formal computational procedure is presented that incorporates information of experimental uncertainties into the surrogate description. It is shown that surrogates then describe a feasible region and are hence not unique. Both physical and chemical properties are utilized as combustion property targets (CPTs) and consistency with experimental formulations is demonstrated for JP-8 and Jet-A (POSF 4658) surrogates. In addition, the use of convex optimization puts existing concepts for surrogate representation on a more rigorous basis and several conclusions are drawn, particularly on the importance of specific CPTs and weighting factors of regression-based approaches. Also, the effect of using simplified models for the evaluation of CPTs on the final surrogate composition is shown by considering the example of linear blending rules for ignition delay. Finally, the surrogate representation problem is connected to multi-parametric optimization and bounds on surrogate compositions are calculated as a function of the experimental uncertainty along with comparisons against experimental results.