(344b) Accelerated Development of Continuous Flow Synthesis of APIs with Multi-Objective Bayesian Optimization Using Automated Dynamic Experimentation

Continuous flow synthesis of pharmaceuticals is attractive due to its small physical footprint, safety, and consistent product quality with precise control of reaction conditions. The lab-scale development of flow API/intermediate synthesis involves optimization of reaction conditions such as reaction temperature, residence time, and stoichiometry, which are traditionally achieved with multiple experiments analyzed at the steady-state. This can be time-consuming and may consume considerable amounts of materials. Herein we present data-rich experimentation to optimize chemical reactions by means of data-driven experiments. Machine learning (ML) guided Bayesian optimization method is developed for continuous tubular reactors operated under dynamic conditions. We designed in-house LabVIEWâ„¢ Virtual Instrument (VI) automation software to control various equipment including Polar Bear Plus Flow^TM reactor (Uniqsis), feed pumps, and in-line Agilent UHPLC. In addition, we enabled automated multi-dimensional sinusoidal dynamic experimentation in the LabVIEWâ„¢ VI.

We present various case studies of automated multi-objective Bayesian optimization using dynamic experimentation for the continuous flow synthesis of APIs at Pfizer as a part of our Flexible API Supply Technology (FAST) effort. The automated sinusoidal dynamic experiment is performed in PFR to collect data of reaction conditions (e.g., residence time and reagent equivalents). The automated sampling of the design space over a parameters trajectory is achieved with Agilent 1290 Infinity II UHPLC system. We also used flow-NMR to identify impurities and track the product concentration. The information obtained for the dynamic experiment is used for both data-driven optimization and chemical kinetics analysis. We will discuss benefits of Bayesian optimization using automated dynamic experimentation for acceleration of development timelines, while reducing the amounts of raw materials consumed, compared to the steady-state approach.