"Machine learning in chemistry and beyond" (ChE-651) seminar by Anirudh Nambiar "Bayesian Reaction Optimization on a Robotic Flow Synthesis Platform"

Machine assistance has helped automate and accelerate steps in the synthesis of organic compounds, accelerating the discovery and development of new medicines and materials. During reaction development, the design space can consist of both continuous (e.g., time, temperature) and categorical (e.g., reagent choice) reaction variables that must be tuned to optimize a desired objective function (e.g., yield). This typically time- and labor-intensive task can be accelerated by leveraging automated synthesis platforms orchestrated by optimization algorithms which efficiently navigate the reaction design space.

This talk will describe my Ph.D. work where we developed a robotic flow synthesis platform with integrated process analytical technology (PAT) to measure reaction outcomes directly on the system. Closed-loop experimentation was established through automated feedback of experimental results to a Bayesian optimization algorithm that suggested which experiment to run next based on prior data.

The first case study will focus on a multi-step synthesis of a small molecule API where the route was proposed by a computer-aided synthesis planning software. Both continuous and categorical variables suggested by the software were optimized using the Bayesian algorithm with respect to multiple objectives simultaneously. In the second case study, Bayesian optimization helped identify optimal continuous variable settings for a photochemical transformation. By generating response surfaces using the algorithm’s mathematical model, the input-output relationship learned by the algorithm will be visualized.