QSPR models to predict the physical hazards of mixtures: a state of art

Mixture Flash Point Calculation: Recent Advances and a Closer Look at Biodiesel

“…24 Commonly used molecular descriptors tend to be categorized as constitutional, electronic, topological, geometrical, thermodynamic, and quantum-chemical. 25,115 Frequently, complex quantum-chemical descriptors are obtained from the detailing of molecular structures with the aid of Density Functional Theory (DFT) modeling. 26 Other simpler approaches to obtain less complex molecular descriptors, such as constitutional ones, are also found in the literature though, such as the use of SMILES 116 or the Simplex Representation of Molecular Structures.…”

“…These authors point out the increasing interest in the development of QSPR based models for FP prediction of mixtures as these tend to be more relevant for industrial purposes. The recent review by Fayet et al 25 focuses on the published literature on mixture properties related to fire hazards. Interestingly both Jiao et al and Fayet et al call attention to the potential of pairing QSPR descriptors to machine learning techniques to obtain more accurate and powerful FP prediction models, which again depends on the availability of FP data, as machine learning algorithms highly depend on training data.…”

“…Due to this fact, a considerably large reliable experimental FP data set is needed. 25 As QSPR initially was developed for the estimation of pure component properties, a few authors have tried to adapt this technique to FP prediction of mixtures; however, some challenges must be addressed when mixtures are being evaluated. 26−30 Usually, mixing rules are employed in order to make possible the development of chemical descriptors in terms of binary mixtures.…”

“…For this reason, initially, they have been successfully applied to predict the FP of pure compounds. , In a traditional approach, authors usually split the experimental FP data set into two parts: the first part is used for training, and the second part is related to external validation of the QSPR model. Due to this fact, a considerably large reliable experimental FP data set is needed …”

“…This lack of experimental data impacts the reliability of QSPR models and influences their performance through propagation effects of uncertainties. 25 Regarding ternary mixtures, most of the studies are relatively recent, except by the study conducted by Saldana et al 31 For ternary or multicomponent mixtures, the task is even more difficult, since the experimental FP data sets available in the literature present poor chemical diversity of mixtures. As mentioned by Fayet and Rotureau, 25 most of the experimental data available targeted a few classes of compounds (hydrocarbons, alcohol, ketones, or acids), which makes the QSPR models limited.…”