(377c) Algorithmic Self-Optimization of Processes in Continuous Flow

Finding the ideal operating conditions of a chemical process has traditionally demanded manual and iterative laboratory work. Researchers will usually apply the one-variable-at-a-time (OVAT) approach until presumably ideal conditions have been found. Recent literature however, describes the use of complex strategies for process optimization, such as the design of experiments (DoE) approach, as well as algorithms from the field of mathematical optimization.¹ Combining such optimization strategies with automated laboratory equipment has allowed for “self-optimizing” chemical systems, also called “self-driving labs”² or more generally, “closed-loop optimization”. These systems aim to find the ideal set of parameters to optimize a desired process step outcome, such as the yield of a chemical reaction, or the particle size for a crystallization step.

The advantages of utilizing these self-optimizing systems, when compared to the OVAT approach, include the reduction of manual laboratory work for researchers, the ability to simultaneously optimize multiple parameters and outcomes at the same time, which saves time and resources, and the increased certainty to find the true global optimum of a process step by leveraging advanced mathematical strategies.

However, due to the highly specialized nature of chemical procedures, the implementation of automated closed-loop optimization systems is a challenging endeavor.³ Additionally, describing a chemical process as an optimization problem in mathematical terms, and subsequently finding the right algorithm for the problem at hand, has proven to be challenging in and of itself.

In this project, we strive to address the difficulties associated with self-optimizing systems laid out above. In our first work, we have established a laboratory setup for the automation and optimization of a biocatalytic reaction in flow via DoE, utilizing 3D printing of ceramics and enzyme immobilization.⁴ Following this, we have worked on the benchmarking of different optimization algorithms, with the inclusion of currently underutilized algorithms, using both theoretical test problems, as well as a Suzuki coupling reaction in continuous flow.⁵ Finally, we are applying the principle of self-optimization to the continuous crystallization of active pharmaceutical ingredients (APIs), via the utilization of 3D printing, optical particle analysis and Bayesian optimization, which has, to our knowledge, not been done before.

Overall, we have been able to advance the state of knowledge in the field of self-optimization by developing new automated setups in continuous flow in the areas of biocatalysis, homogeneous catalysis, and continuous crystallization of APIs, as well as further the understanding of algorithm usage for real-life process optimization.

1. Mateos, C., Nieves-Remacha, M. J. & Rincón, J. A. Reaction Chemistry and Engineering 4, 1536–1544 (2019).
2. Abolhasani, M. & Kumacheva, E. Nat. Synth 1–10 (2023).
3. Christensen, M. et al. Chemical Science 12, 15473–15490 (2021).
4. Valotta, A. et al. Journal of Flow Chemistry 11, 675–689 (2021).
5. Soritz, S., Moser, D. & Gruber-Wölfler, H. Chemistry–Methods 2, (2022).