Combining Theoretical and Data-Driven Approaches To Predict Drug Substance Hydrate Formation

Hydrates represent a very significant fraction of pharmaceutical molecular crystals and can be leveraged to simplify downstream processing for formulations such as wet granulation and hot-melt extrusion. In silico methods to predict hydrate formation can guide experimental screening and evaluate residual risk of selected forms. Both solution mixing thermodynamics and relative propensities of hydrogen bond formation can be used for virtual screening. Our study assessed these techniques for a previously studied set of relatively simple drug compounds (average molecular weight 300) and a new set of more complex AbbVie-pipeline compounds (average molecular weight 550). Although solution thermodynamics have been shown to successfully discriminate hydrate formation for the set of smaller drug molecules, this technique did not provide successful screening for the more complex set of AbbVie compounds tested. A single-differential hydrogen bond propensity (SD-HBP) score, which accounts for only the strongest donoracceptor pairing in both anhydrate and hydrate forms, also provides little utility. We therefore developed a multidifferential hydrogen bond propensity (MD-HBP) score that considers the competitive effect of multiple donoracceptor interactions in each form. Additionally, the MD-HBP score utilizes solid-state conformations (estimated here via COSMO-RS theory) to strengthen the data-driven analysis of the solid-state and ensure more accurate description of possible hydrogen bond networks in anhydrate and hydrate solid forms. This quantitative MD-HBP score performed well at differentiating between hydrate-forming and non-hydrate-forming compounds for both sets of compounds; thus, it can be applied more broadly in solid form development.