Assessing Solvate Prediction Approaches: A Case of Spironolactone

In silico methods for predicting solvate formation can guide and simplify solid form screening and solvent selection in crystallization processes. Solvates involve many complicated nonspecific interactions, making solvate prediction challenging. The applicability of three solvate prediction approaches to spironolactone (SPI) in 29 solvents was assessed, including thermodynamic excess enthalpy (Hex), hydrogen bond propensity (HBP), and a random forest model (RF). In experimental screening, 6 new solvates were identified and crystal structures of 5 solvates were revealed. Structural features, interaction calculations, and thermal analysis provided insights into the formation of SPI solvates. Not only strong hydrogen bonds but also weak interactions contribute significantly to the SPI channel and isolated site solvates. The Hex based on COSMO-RS theory (conductor-like screening model for real solvents) has a 65.5% success rate for SPI solvates prediction. The HBP can only be used for solvents with hydrogen bond donors with an 81% hit rate. The RF model shows a predictive success rate of 76.9% for unreported solvents. This work indicates it is difficult for these models to adequately predict solvate systems where multiple dominant interactions (such as hydrogen bonds, van der Waals forces, etc.) may be involved simultaneously. Using both COSMO-RS and HBP methods followed by the RF model may yield better results.