Survey of the solubility data of organic molecules. Classification and modeling of the factors influencing Setschenow coefficients in several ionic media

In this contribution, a survey of the total and specific solubility of several organic molecules is performed. Even if the importance and the impact of the knowledge of the solubility is obvious (not only for the field of analytical chemistry), the experimental determination of these parameters, and of the Setschenow coefficient1 is often limited to few articles per year, and consequently the number of indexed scientific journals interested on these topics is generally decreasing. As an example, no database are available for this kind of information. For this purpose, a quantitative analysis of the factors influencing the solubility and the modeling of the Setschenow coefficient can be helpful to rationalize the results published in the past, to give tools to estimate the solubility of new compounds, and predict the Setschenow coefficient, the parameter that accounts for the increase (salting-in) or decrease (salting-out) of the solubility when the concentration of the background salt increases. Literature data, relative to total and specific solubility on pure water, together with Setschenow coefficients in different ionic media, were collected and critically analyzed, assigning a weight on the basis of the reliability of the experimental determination. Then, each datum was assigned to a class, based on chemical properties, such as structure (linear or cyclic), functional group (carboxylates, amines or mixtures), number of carbon atoms, values of the protonation constants etc… Once the literature data are rationalized and classified, they can be modeled using empirical equations and principal component analysis to obtain quantitative information about the descriptors of the Setschenow coefficients and the solubility in pure water. The results will be tested to check the goodness of the fit and the models will be validated splitting the sample in a test set, containing 90% of the whole data, and an evaluation set, containing the remaining 10% of the sample. The models obtained for the estimation of the Setschenow coefficients will be compared with the existing ones, such as those of Yu et al. (based on IEF-PCM calculations)2, Xu et al. (based on QSPR studies)3 and Ni and Yalkowsky (based on water/octanol partition coefficient)4.