2023 AIChE Annual Meeting

(628a) Application of Experimental Design Via Bayesian Optimization (EDBO) to Pharmaceutical Process Characterization and Control Strategy Design

Authors

Jose Tabora - Presenter, Bristol-Myers Squibb Company
Michael Williams, Bristol Myers Squibb
Jason Stevens, Bristol Myers Squibb
Jiayi Fu, Bristol Myers Squibb
Jun Li, Bristol-Myers Squibb
Damian Reyes-Luyanda, Bristol Myers Squibb
Dimitri Skliar, Bristol-Myers Squibb
Previous studies have demonstrated the applicability of Experimental Design via Bayesian Optimization (EDBO) to chemical synthesis and chemical processes. In addition multiple research groups have published software that facilitates the programmatic incorporation of EDBO into a research organization [1,2].

In pharmaceutical development, the late-stage process characterization workflow, is enabled by the application of Design of Experiments (DOE) which allow the identification of a robust control strategy that takes into account the multivariate nature of the underlying unit operation. Typically, an experimental design of a targeted optimality criteria is performed from which models are built to estimate the level of the responses (critical quality attributes, cQA’s). The quantification of the responses allows the construction of a design space which provides assurance of control.

In this work we demonstrate that applying EDBO to the characterization workflow results in an efficient and streamlined workflow for control strategy design. We demonstrate the application of the procedure in two instances of reaction characterizations with multiple factors and responses. We propose a methodology to seed the algorithm with an appropriate subset of the DOE design to improve the efficiency of process characterization. Finally we suggest a visualization strategy to improve the interpretability of the surrogate process model to facilitate the design of a multivariate control strategy.

[1] Wang Y, Chen TY, Vlachos DG. NEXTorch: a design and Bayesian optimization toolkit for chemical sciences and engineering. Journal of Chemical Information and Modeling. 2021 Oct 25;61(11):5312-9.

[2] Torres JA, Lau SH, Anchuri P, Stevens JM, Tabora JE, Li J, Borovika A, Adams RP, Doyle AG. A Multi-Objective Active Learning Platform and Web App for Reaction Optimization. Journal of the American Chemical Society. 2022 Oct 19;144(43):19999-20007.