A Multilayer Network Approach for Predicting the Cocrystal–Salt Transition

Multicomponent crystals (MCCs) are an important tool for improving medicines and other useful chemicals. However, finding compounds that can cocrystallize is time-consuming and expensive. Computational methods can be used to speed up the process of discovering cocrystallizing compounds. Link prediction, which uses a network of known MCCs to predict new MCCs, is such a method. It has previously been used to predict cocrystals and solvates. In this work, we show that link prediction can also be used to predict salts. Moreover, by creating a multilayer network of cocrystals and salts, it is possible to predict whether a given pair of chemicals will form a cocrystal or a salt. We demonstrate that the Δ pK a rule, which has previously been used to distinguish cocrystals and salts, is implicitly included in this multilayer salt–cocrystal network. With the multilayer network approach, link prediction can even be used to distinguish cocrystals and salts in the Δ pK a range between −1 and 4, where the Δ pK a rule itself is inconclusive.