Vapourtec Ltd
Telephone +44(0)1284 728659
Email [email protected]
Photoredox catalysis is recognized for its important role in the synthesis of pharmaceutically relevant compounds such as C(sp³)-rich tertiary amines. However, the development of robust processes has been challenged by the difficulty in identifying underlying mechanistic models and the large reaction space associated with this class of transformations. In this work, a machine learning-driven optimization of a photoredox tertiary amine synthesis was carried out using six continuous variables (e.g., concentration, temperature, residence time) and solvent choice as a discrete variable, within a semiautomated continuous flow setup. A large library of solvents was initially considered, and the discrete space was narrowed through multiple steps of a priori knowledge generation, including solubility predictions.
Subsequently, a novel Bayesian optimization algorithm, nomadic exploratory multiobjective optimization (NEMO), was deployed to identify and populate the Pareto front for the dual reaction objectives—yield and reaction cost. The most influential parameters for achieving high yield were identified using permutation feature importance and partial dependence plots. These included sig3, representing the asymmetry of the s-profile in the discrete space, as well as the equivalences of alkene and Hantzsch ester in the continuous space. Catalyst loading and residence time were found to be correlated with absorbed photon equivalence, while catalyst loading was also identified as the primary factor influencing cost. Although productivity was not directly targeted as an optimization objective, a ∼25-fold improvement over batch reactions was achieved in flow, corresponding to a throughput of approximately 12 grams per day.
- Jorayev, P
- Soritz, S
- Sung, S
- Jeraal, MI
- Russo, D
- Barthelme, A
- Toussaint, FC
- Gaunt, MJ
- Lapkin, AA
- Department of Chemical Engineering and Biotechnology, University of Cambridge, Cambridge CB3 0AS, United Kingdom
