Structured HTE Data and Integrated Bayesian Optimization Accelerate Scalable Amidation Process Development
Jason M. Stevens, Dung L. Golden, Nikki Dare, Shubhangi Aggarwal, David Del Valle, Stacey DiSomma, Michael Eng, Ariel Furman, Lauren Grant, Diane Hendrix, Bilal Hoblos, Jiajie Huo, Mikalai Malashchonak, Ketleine Miller, Emily Mumford, Cheng Peng, Lingxiang Lu, Donggeon Nam, Frederick Roberts, Victor Rosso, Eric Saurer, Sushant Singh, Dmitriy Snetkov, Anton Terekh, Alexander E. Waked, Michael Williams, Chengmin ZhangAbstract
We describe the evolution of our early hit-identification workflow for reagent-mediated amidation reactions in support of commercial route development. Over the past decade, our strategy progressed from broad, exhaustive high-throughput experimentation (HTE) with limited data maturity to a modern, enterprise-scale digital platform that embeds data science tools directly into the experimental workflow. By tailoring Katalyst D2D─a commercial electronic laboratory notebook (ELN) for HTE developed by ACD/Laboratories─we established a structured, machine-readable data architecture and enabled seamless integration with the EDBO+ Bayesian optimization engine. Retrospective analysis of historical amidation data sets showed that, despite screening more than 60 reagents, only a small subset consistently translated to commercial processes. This insight motivated a streamlined workflow in which chemists begin with a focused survey of six reaction conditions rooted in empirical performance and commercial precedent, followed by Bayesian optimization within the ELN to identify optimal reagent/base/solvent combinations. This study highlights how treating reaction data as a digital asset─FAIR, interoperable, and directly linked to optimization tools─accelerates the transition from reaction screening to robust process development.