Evaluation of Predictive Solubility Models in Pharmaceutical Process Developmentan Enabling Technologies Consortium Collaboration

Solubility is an important property in the design of
many unit
operations within the development of synthetic pharmaceutical processes.
However, the direct measurement of solubility in potential solvents
and solvent mixtures is not always ideal or practical. Several models
have been developed and gained usage within the pharmaceutical industry
to predict the solubility of compounds within organic solvents. As
part of an Enabling Technologies Consortium (ETC) collaboration, the
need for pre-competitive benchmarking was identified to guide the
use of solubility models. In this paper, a set of predictive solubility
models are evaluated against an extensive set of measured solubilities
(24 solutes over 80 solvent and solvent mixtures). Acetazolamide,
caffeine, cimetidine, cinnarizine, glyburide, omeprazole, paracetamol,
piroxicam, risperidone, and saccharin were measured as part of a joint
effort within ETC. Additional in-house solubility data, spanning a
more limited range of solvents, were contributed by ETC collaboration
participants. The solubility models vary in their approach from first-principles
models (COSMO-RS and COSMO-SAC) to regression-based models (R-UNIFAC
and the “Lovette–Amgen” model). The performance
of each model was extensively evaluated using different fitness measures.
This benchmarking study supports the continued application of solubility
modeling as a tool to augment experimentation within chemical process
development.