Biocatalysis represents an invaluable tool for developing more sustainable pharmaceutical manufacturing processes. However, the optimisation of these complex transformations remains challenging when using traditional methods. Recent advances in the field of chemical reaction optimisation have employed Bayesian optimisation, which has been successfully integrated into automated flow reactors to develop self-optimising reactor platforms. Although self-optimisation has been applied to chemical transformations with great effect, it has not yet been applied to biocatalytic reactions. In this work, we report the first application of multiobjective and mixed variable Bayesian optimisation for the automated development of a biocatalytic transformation. Using this approach, we develop a significantly improved process for the direct amidation of a β-ketoester in just 31 hours of experimental time, while also extracting insights from the black-box models to understand solvent-dependant effects and interactions among the reaction conditions.
