Premium
New QSPR Models to Predict the Flammability of Binary Liquid Mixtures
Author(s) -
Fayet Guillaume,
Rotureau Patricia
Publication year - 2019
Publication title -
molecular informatics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.481
H-Index - 68
eISSN - 1868-1751
pISSN - 1868-1743
DOI - 10.1002/minf.201800122
Subject(s) - quantitative structure–activity relationship , flash point , multilinear map , ternary operation , binary number , applicability domain , molecular descriptor , series (stratigraphy) , chemistry , thermodynamics , biological system , mathematics , computer science , organic chemistry , stereochemistry , paleontology , physics , arithmetic , pure mathematics , biology , programming language
New Quantitative Structure‐Property Relationships (QSPR) are presented to predict the flash point of binary liquid mixtures, based on more than 600 experimental flash points for 60 binary mixtures. Two models are proposed based on a GA‐MLR approach that uses a genetic algorithm (GA) variable selection in multilinear regressions (MLR). In these models, mixtures were characterized by a series of mixture descriptors calculated from various mixture formula combining the molecular descriptors of the single compounds constituting the mixtures and their respective molar fractions in the mixture. The best model demonstrated good predictive capabilities with a mean absolute error of only 7.3 °C estimated for an external validation set. Moreover, this model is focused on mixture descriptors applicable to more complex mixtures, i. e. constituted of more than 2 components, and already demonstrated interesting predictions for a series of ternary mixtures.