Bayesian optimization for chemical reactions
Stefan Desimpel, Matthieu Dorbec, Kevin M. Van Geem, Christian V. Stevens
Abstract
Bayesian optimization (BO) enables data-efficient optimization of complex chemical reactions by balancing exploration and exploitation in large, mixed-variable parameter spaces. This review provides an accessible introduction for chemists wishing to adopt BO, outlining the fundamentals of surrogate models, acquisition functions, and key mathematical concepts. Practical considerations are emphasized, including kernel design, representation of categorical variables, and strategies for multi-objective and batch optimization. Applications are comprehensively surveyed across experimental scales, from high-throughput platforms to automated flow reactors and larger-scale processes. Finally, emerging directions such as transfer learning and data reuse are discussed in the context of accelerating optimization campaigns and enabling more generalizable, data-driven strategies in chemistry.