(667f) Automated in silico Crystallization Process Design Using Solubility Models; Web Applications for Visualization and an Overview of a Solvent Selection Workflow

The accuracy of solubility models for realistic active pharmaceutical ingredients (API) has improved to the point that qualitative and quantitative predictions in aqueous and organic solvents are now becoming routine. In crystallization process design, such solubility predictions can be leveraged as a first step for selection of realistic solvent systems or isolation method (e.g. thermal, antisolvent or reactive crystallization) before going into the laboratory. In this contribution, a workflow is presented to show how such predictions are leveraged to aid in crystallization process design at Lilly. Additionally, custom Python and R code is described that takes in the modeled solubility data and simulates ~10000 possible crystallization processes for consideration as a starting point for process design. A user interface built upon web applications (Bokeh and Rshiny) for the visualization and process selection is also described. Application examples including R-UNIFAC from dynochem and Cosmotherm solubility models will be discussed.