(446e) Optimizing Multistep Continuous Flow Organic Synthesis with Bayesian Optimization and Robotics

Continuous flow synthesis with multiple unit operations (reactors, separators) telescoped in series has enabled multistep assembly of complex molecules under intensified process conditions with minimal manual purification between steps [1]. Automated experimentation platforms orchestrated by intelligent algorithms capable of optimizing reaction conditions towards a desired objective have further reduced the manual burden during process development [2]. While the majority of prior work on flow reaction optimization has focused on single-step reactions, recent demonstrations involving sequential reaction optimization [3], reaction analysis at multiple points [4], and multi-objective optimization of a telescoped reaction-separation sequence [5] have started to address certain aspects of multistep reaction optimization.

In this work, we present a robotically reconfigurable flow chemistry platform for executing, analyzing, and optimizing multistep flow organic syntheses. The platform contains a library of process modules that can be robotically placed in any order onto a process stack for performing reactions (in tubular reactors with different volumes and a packed bed), membrane-based phase separations, and inline analysis (FT-IR and LC-MS). The hardware is coupled to a Bayesian optimization algorithm capable of handling continuous variables (e.g., temperature, residence time) and categorical variables (e.g., reagent type), as well as optimizing multiple objectives simultaneously (e.g., yield, productivity).

Using an exemplary multistep synthesis of a small-molecule drug, we demonstrate the following: (a) optimization of all reactions in a multistep synthesis simultaneously (as opposed to individually) for identifying globally optimal conditions, (b) inline analysis at multiple locations along a multistep sequence (as opposed to only at the end) for quantitatively assessing how the overall yield is affected by individual yields, and (c) robotic reconfiguration of reactor volume provides an additional degree of freedom that enables variation of downstream residence times which are constrained by upstream flow rates in multistep flow processes.