If optimizing for general parameters in chemistry is useful, why is it hardly done?

Stefan P. Schmid, Ella Miray Rajaonson, Cher Tian Ser, Mohammad Haddadnia, Shi Xuan Leong, Alan Aspuru-Guzik, Agustinus Kristiadi, Kjell Jorner, Felix Strieth-Kalthoff

Submission Track: Full Paper

Submission Category: AI-Guided Design + Automated Synthesis

Keywords: Bayesian Optimization, Generality, Transferable Optima, Reaction Conditions, Condition Optimization

TL;DR: This work benchmarks the effectiveness of Bayesian optimization in discovering general and transferable optima, at the example of chemical reaction optimization.

Abstract: General parameters are highly desirable in the natural sciences — e.g., reaction conditions that enable high yields across a range of related transformations. This has a significant practical impact since those general parameters can be transfered to related tasks without the need for laborious and time-intensive re-optimization. While Bayesian optimization (BO) is widely applied to find optimal parameter sets for specific tasks, it has remained underused in experiment planning towards such general optima. In this work, we consider the the real-world problem of condition optimization for chemical reactions to study whether performing generality-oriented BO can accelerate the identification of general optima, and whether these optima also translate to unseen examples. This is achieved through a careful formulation of the problem as an optimization over curried functions, as well as systematic benchmarking of generality-oriented strategies for optimization tasks on real-world experimental data. We find that the optimization for general reaction conditions are determined by the sampling of substrates, with random selection outperforming more data-driven strategies.

Submission Number: 86