(15f) Closed-Loop Reaction Optimization in Microscale Oscillating Droplets: An MINLP Algorithm Applied to Suzuki-Miyaura Coupling Catalyst Selection
- Conference: AIChE Annual Meeting
- Year: 2018
- Proceeding: 2018 AIChE Annual Meeting
- Group: Pharmaceutical Discovery, Development and Manufacturing Forum
- Time: Sunday, October 28, 2018 - 5:10pm-5:30pm
The optimization of reaction conditions can help achieve this goal, but the inherent combinatorial nature of selecting reaction conditions (e.g., catalyst identity, concentration, reaction time, temperature) means that performing an exhaustive screen is counter to the goal of efficiency, both in terms of material usage and time. More focused experimentation is possible through the use of optimization algorithms that propose experiments based on accumulated data, so that only useful experiments need be performed.
Building on previous work in the group , here we report a mixed-integer nonlinear program (MINLP) algorithm for the simultaneous optimization of both discrete and continuous variables. The algorithm consists of two phases (1) an initial D-optimal design, which proposes a diverse range of experiments to regress a quadratic response surface model, and (2) an iterative branch-and-bound approach to discrete variable selection, where G-optimal experiments are proposed to minimize the variance at predicted optima. The algorithm is designed to optimize catalyst turnover number (TON) subject to a minimum yield constraint.
We first validate its performance using a suite of five simulated test cases representing variations on a catalytic bimolecular reaction. Next, we validate its performance experimentally using the exemplary Suzuki-Miyaura cross-coupling of 3-chloropyridine and 2-fluoropyridine-3-boronic acid pinacol ester. Experimental validation was conducted using an automated chemistry platform that performs reactions sequentially at the 20 microliter scale [2,3]. Yield and TON are calculated via online HPLC/MS and automatic peak integration. This closed-loop optimization proceeds without any human intervention. The new algorithm exhibits significantly faster convergence than the previous version and is found to be robust to measurement error.
 B. J. Reizman, K. F. Jensen, Accounts of Chemical Research 2016, 49, 1786-1796.
 Y.-J. Hwang*, C. W. Coley*, M. Abolhasani, A. L. Marzinzik, G. Koch, C. Spanka, H. Lehmann and K. F. Jensen, Chemical Communications, 2017, 53, 6649â6652.
 C. W. Coley, M. Abolhasani, H. Lin and K. F. Jensen, Angewandte Chemie, 2017, 129, 9979â9982.