Solubility of Sulfanilamide and Sulfacetamide in neat solvents: Measurements and interpretation using theoretical predictive models, first principle approach and artificial neural networks

Solubilities of Sulfanilamide and Sulfacetamide in neat solvents were both measured and collected from the literature. The set comprising 35 systems was interpreted in terms of the empirical and semi-empirical predictive models including Buchowski-Ksiazczak model (lambda h-equation), extended Buchowski's model (lambda beta-equation), modified Apelblat equation, van't Hoff-Yaws model, Non-Random Two Liquid (NRTL) model, Wilson model and the Weibull two-parameter extrapolation model. The advantage of corrected Akaike information criterion was used for precise quantification of the quality of each model. It turned out that the lowest values of this measure were found for the Buchowski-Ksiazczak approach indicating the best correlation with experimental solubility data. Interestingly, lambda beta-equation and van't Hoff-Yaws model give a fitting of almost the same quality but were slightly less accurate compared to lambda h-equation. The results of applied COSMO-RS computations suffered seriously from inaccuracies providing only a qualitative guess of solubility. Hence, machine learning procedure was applied for building a non-linear fully predictive model. The set of used molecular descriptors comes from COSMO-RS computations and characterizes intermolecular interactions in pure solvents. Based on the quality of obtained machine learning model it is possible to confirm that the selected set of descriptors can be used for solubility modeling and carries the most important information necessary for quantifying of saturated solutions. (C) 2020 Elsevier B.V. All rights reserved.

Automated summary

The solubilities of Sulfanilamide and Sulfacetamide in various solvents were studied using different predictive models. The Buchowski-Ksiazczak approach showed the best correlation with experimental data, while machine learning was used to develop a more accurate predictive model.