Property Prediction of Pharmaceuticals for Designing of Downstream Separation Processes

Downstream processing of pharmaceuticals is often a complex and expensive process section in pharmaceutical industry. The separation and purification cost of small to medium-sized active pharmaceutical ingredients (APIs) can be up to 70-90% of the total production cost. Designing a downstream processing includes selection, arrangement and rigorous evaluation of various unit operations such as crystallization and extraction.

Solvent screening and selection is a critical step in designing such unit operations. The solubility of APIs is a major property for solvent screening and selection. Solubility can either be determined experimentally or by using predictive thermodynamic models. However, experimental solvent screening is most often expensive and laborious, or sometimes simply not possible due to the lack of sufficient amount of an API.

Moreover, designing some unit operations; for example, crystallization further requires solubility property of an API as a function of temperature, solvent composition and pH for cooling, antisolvent and reactive crystallization, respectively, or any combination thereof. In this work, different predictive thermodynamic models such as COSMO-SAC, COSMO-RS, NRTL-SAC and UNIFAC were applied for solubility prediction of a large molecular weight steroid-like structure antibiotic.

The predicted solubility data were compared with experimental solubility data in a set of various solvents representative of wide solvent properties. Therefore, a model which gives more accurate solubility prediction can be selected. For this specific case study, the root mean square errors (RMSEs) of COSMO-SAC, UNIFAC, COSMO-RS and NRTL-SAC are 54.3%, 51.8%, 34.3% and 4.8%, respectively.

NRTL-SAC was selected as a property prediction tool for solvent screening and selection as well as for further designing and optimization of crystallization unit operation. Moreover, global sensitivity and uncertainty analyses of the predictive thermodynamic model parameters were performed in order to identify sources of uncertainty and evaluate their influence.