Application of Multivariate Adaptive Regression Splines (MARSplines) Methodology for Screening of Dicarboxylic Acid Cocrystal Using 1D and 2D Molecular Descriptors

摘要

Dicarboxylic acids (DiAs) are probably among of the most popular cocrystal formers. Due to the high hydrophilicity and nontoxicity, they are promising solubilizers of active pharmaceutical ingredients (APIs). Although DiAs appear to be highly capable of forming multicomponent crystals with various compounds, some systems reported in the literature are physical mixtures of the solid state without forming stable intermolecular complex. In this study, an accurate cocrystal screening model was developed based on the MARSplines (Multivariate Adaptive Regression Splines) methodology and easily computable descriptors driven simply from the SMILES codes. Additionally, the data set was enriched with several new mixtures of sulfamethazine. As demonstrated, this sulfonamide can form new multicomponent crystals with oxalic, malonic, and maleic acids. In the case of the latter system, a significant 10-fold solubility advantage was observed. The whole data set comprised 608 cocrystals and 104 systems hardy miscible in the solid state, denoted as simple eutectics. The final 7-factor equation was subjected to external and internal validation procedures, which indicated its high predicting power. The reliability of the proposed approach can be illustrated by the proper classification probability of cocrystals reaching 91%. The classification quality of simple binary eutectics was found to be only slightly worse (TN% = 81%).