(198l) Application of SAFT-? Mie to Solubility Prediction and Solvent Selection

Solubility is a critical physical property of drug molecules in development with a significant impact across all the Pharmaceutical Sciences disciplines. Chemical process optimisation, formulation design, and the bio-performance of the product all highly depend on the solubility of the solute in question. Identifying the optimal solvents at different stages of drug development is a challenging task, and one that is difficult to tackle with experimental measurements alone. Several thermodynamic models have been used in this context, varying from approaches that rely more on experimental data, such as the NRTL-SAC and PC-SAFT methods, to more predictive models, e.g., the COSMO-RS and UNIFAC methods.

Group-contribution (GC) methods combined with molecular based theories, such as the SAFT family of equations of state, are particularly suitable for the application of solubility prediction. Such models combine the predictive accuracy of the GC framework with the range of applicability of the underlying thermodynamic model. SAFT-type equations of state have been successfully applied to the modelling of a wide range of systems, from polymers to highly associating compounds and mixtures of electrolytes. As an example of such an approach, in this work we present the application of the SAFT-γ Mie equation of state to solubility modelling.

We will present how SAFT-γ Mie can be used in a fully predictive fashion to the solubility modelling of solutes when parameters for all the relevant functional groups are available, by examining a number of industry-relevant examples. Results across a range of solvents will be presented and comparisons to the results obtained by other commercially available methods will be made. In addition to this we will demonstrate how solubility data can be used to improve the model predictions, but also to develop SAFT-γ Mie models for solutes that cannot be fully described by the available functional groups. Different modelling strategies will be presented, and the resulting predictive capabilities will be compared.