(56c) Determining Solubility of APIs and Intermediates From Automated Parallel Experiments and Modeling

The design of efficient pharmaceutical processes such as reaction and crystallization is critically dependent on choosing an optimal solvent system.  Selecting the solvent system requires the knowledge of the solubility of APIs or intermediates in several solvents and solvent mixtures.  However, even with the use of automation and material sparing techniques, the direct experimental measurement of solubility in all possible solvents and solvent mixtures and as a function of temperature is often impractical and inefficient due to the limited quantities of material available during the early stages of process development.  To address this issue, earlier we have presented an approach based on combined automated parallel experiments and modeling to estimate the solubility in a large number of solvents and solvent mixtures.  Here we explore the value of different solubility models, their experimental input requirements, output capabilities, and the overall success of an integrated empirical-modeling solubility workflow.  The solubility modeling is typically performed using the NRTL-SAC (Nonrandom Two-Liquid Segment activity coefficient) model [1].  We compare the performance of NRTL-SAC with two other alternative models: PC-SAFT (Perturbed-Chain Statistical Associating Fluid Theory) [2] as implemented by Scienomics and regressed-UNIFAC (UNIversal quasichemical Functional-group Activity Coefficients) as implemented by Dynochem.  Unlike the standard UNIFAC [3], the regressed-UNIFAC [4] from Dynochem treats the solute as a single functional group.  Furthermore, we evaluate the predictive capabilities of these three different models and discuss their unique strengths, limitations, and accuracy.  We also discuss the FTE (Full Time Equivalent) savings obtained through our combined automated parallel experiment and modeling approach in the pharmaceutical process design. 

References

---