Prediction of Solvents Suitable for Crystallization of Small Organic Molecules

Abstract

Selection of suitable solvent is essential for crystallization of pharmaceuticals. Based on chemical structures of 6397 compounds and 15 single solvents that were used to obtain their single crystals, correlations between the molecular characteristics and the solvents have been investigated by cheminformatics methods. Decision-tree and Bayesian-probability methods have been applied to make classification models. These two models are complementary in character in the present case. It has been proven that the prediction of the solvent rankings for particular compounds by use of the classification models is satisfactory from the practical point of view. The present study has demonstrated that cheminformatics methods could greatly help rational crystallization of small organic molecules such as pharmaceuticals.