The simplex algorithm for the rapid identification of operating conditions during early bioprocess development: case studies in FAb' precipitation and multimodal chromatography

This study describes a data-driven algorithm as a rapid alternative to conventional Design of Experiments (DoE) approaches for identifying feasible operating conditions during early bioprocess development. In general, DoE methods involve fitting regression models to experimental data, but if model fitness is inadequate then further experimentation is required to gain more confidence in the location of an optimum. This can be undesirable during very early process development when feedstock is in limited supply and especially if a significant percentage of the tested conditions are ultimately found to be sub-optimal. An alternative approach involves focusing solely upon the feasible regions by using the knowledge gained from each condition to direct the choice of subsequent test locations that lead towards an optimum. To illustrate the principle, this study describes the application of the Simplex algorithm which uses accumulated knowledge from previous test points to direct the choice of successive conditions towards better regions. The method is illustrated by two case studies; a two variable precipitation example investigating how salt concentration and pH affect FAb' recovery from E. coli homogenate and a three-variable chromatography example identifying the optimal pH and concentrations of two salts in an elution buffer used to recover ovine antibody bound to a multimodal cation exchange matrix. Two-level and face-centered central composite regression models were constructed for each study and statistical analysis showed that they provided a poor fit to the data, necessitating additional experimentation to confirm the robust regions of the search space. By comparison, the Simplex algorithm identified a good operating point using 50% and 70% fewer conditions for the precipitation and chromatography studies, respectively. Hence, data-driven approaches have significant potential for early process development when material supply is at a premium.