Data-Science Driven Autonomous Process Optimization

Abstract

Autonomous process optimization involves the human intervention-free exploration of a range of pre-defined process parameters in order to improve responses such as reaction yield and product selectivity. Utilizing off-the-shelf components, we developed a closed-loop system capable of carrying out parallel autonomous process optimization experiments in batch with significantly reduced cycle times. Upon implementation of our system in the autonomous optimization of a palladium-catalyzed stereoselective Suzuki-Miyaura coupling, we found that the definition of a set of meaningful, broad, and unbiased process parameters was the most critical aspect of a successful optimization. In addition, we found that categorical parameters such as phosphine ligand were vital to determining the reaction outcome. To date, categorical parameter selection has relied on chemical intuition, potentially introducing an element of bias into the experimental design. In seeking a systematic method for the selection of a diverse set of phosphine ligands fully representative of the chemical space, we developed a strategy that leveraged computed molecular descriptor clustering analysis. This strategy allowed for the successful autonomous optimization of a stereoselective Suzuki-Miyaura coupling between a vinyl sulfonate and an arylboronic acid to selectively generate the E-product isomer in high yield.

The materials tetrahedron has a “digital twin”

“…Similar opportunities and challenges for closed-loop automation exist outside the field of materials science as well, such as drug discovery, 67 healthcare, 68 supply chain management, 69 architecture, 70 and chemistry. 71,72 Although an exhaustive review falls outside the scope of this perspective, we urge the interested reader to consult recent review articles and perspectives that explore the intersection of materials science with one or more of the topics in information science outlined in this perspective. These resources include discussions of data ecosystems and infrastructures for metadata management; [73][74][75] high-throughput library generation and characterization; 76,77 successes and challenges of Materials Genome Initiative research, machine learning methods, and computational materials databases; [78][79][80] methods for linking materials characterization and computation across length scales; 81,82 methods, representations, and applications in the field of polymer informatics; 83 materials discovery for energy applications; 84 and integrated systems for next-generation microscopy.…”

A Materials Acceleration Platform for Organic Laser Discovery

“…While traditional materials discovery is a slow process, recent developments in self-driving labs can accelerate the optimization process dramatically by combining automated experimental systems with data-driven workflows. Along with the machine learning revolution, research groups and industries have been developing better and more reliable automated systems for use in the fields of chemistry [14][15][16][17][18][19][20] , biotechnology [21][22][23] and electronics 24,25 . For example, Li et al 26 developed an automated and modular building block-based synthesis platform based on iterative Suzuki-Miyaura cross-coupling reactions.…”

Golem: an algorithm for robust experiment and process optimization

“…In chemistry, these parameters may be the experimental conditions that control the yield of the reaction, or those that determine the cost-efficiency of a manufacturing process (e.g., temperature, time, solvent, catalyst). 1,2 The design of molecules and materials with specific properties is also a multi-parameter, multi-objective optimization problem, with their chemical composition ultimately governing their properties. [3][4][5][6][7] These optimization tasks may, in principle, be performed autonomously.…”

“…, temperature, time, solvent, catalyst). 1,2 The design of molecules and materials with specific properties is also a multi-parameter, multi-objective optimization problem, with their chemical composition ultimately governing their properties. 3–7 These optimization tasks may, in principle, be performed autonomously.…”