Part 1: Development of Bayesian Optimization for Chemical Synthesis
Part 2: Lab Automation Leveraging Bayesian Optimization for Chemical Process Development
Reaction optimization is pervasive in reaction development, catalyst discovery, and in multi-step synthesis of functional molecules. Bayesian optimization has recently emerged as an efficient approach to (hyper)parameter tuning of machine learning models. In part 1, Professor Doyle will discuss the development of the open-source Bayesian optimization software, Experimental Design by Bayesian Optimization (EDBO), on the basis of two published high-throughput experimentation datasets. Performance was assessed on a new HTE dataset for Pd-catalyzed C–H arylation and as compared to chemists’ decision making in reaction optimization via an online game. For part 2, Jay Stevens will present the technical and logistical considerations for deploying EDBO, using laboratory automation. Specifically, the optimization of the Mitsunobu reaction recently reported by Shields (Nature, 2021, 590, 89–96) will be discussed as well as a recent example from the Bristol-Myers Squibb chemical process development portfolio.
Abigail Doyle, PhD
A. Barton Hepburn Professor of Chemistry
Jason Stevens, PhD
Computer Assisted Synthesis Lead