Machine learning-driven optimisation of continuous flow photoredox amine synthesis

Photoredox catalysis has been recognized as playing an important role in the synthesis of pharmaceutically relevant compounds, such as C(sp³)-rich tertiary amines. The identification of underlying mechanistic models for such novel transformations, combined with the large reaction space of this class, has been found to present challenges in the development of a robust process.

In this work, the machine learning-driven optimisation of a photoredox tertiary amine synthesis was demonstrated using six continuous variables (e.g., concentration, temperature, residence time) and solvent choice as a discrete variable within a semi-automated continuous flow setup. A large solvent library was initially considered, and the discrete space was narrowed through several steps of a priori knowledge generation, including solubility predictions.

A novel Bayesian optimisation algorithm, Nomadic Exploratory Multi-Objective optimisation (NEMO), was employed to identify and populate the Pareto front for the dual objectives of yield and reaction cost. The most influential parameters for high yield were identified through permutation feature importance and partial dependence plots, including sig3 (the asymmetry of the -profile for the discrete space) and the equivalences of alkene and Hantzsch ester among the continuous variables. Catalyst loading and residence time were found to correlate with absorbed photon equivalence, while catalyst loading was also identified as the primary driver of reaction cost.

Although productivity was not defined as an optimisation objective, a ~25-fold increase was achieved in flow compared to batch conditions, corresponding to a throughput of approximately 12 grams per day.

• Jorayev, P
• Soritz, S
• Sung, S
• Jeraal, M
• Russo, D
• Barthelme, A
• Toussaint, F
• Gaunt, M
• Lapkin, A

• University of Cambridge

• Request a Brochure
• Request more Info
• Request a Quote