Bayesian Optimization of a Suzuki Coupling for the Industrial Synthesis of a Sartan Drug Intermediate

Abstract

Sartans are an important family of drugs used to treat hypertension. A common characteristic of most of them is a biphenyl moiety, commonly obtained through a Suzuki-Miyaura cross-coupling reaction, of which thousands of tons are produced every year. Bayesian Optimization (BO), a machine-learning technique to find the optimum of complex unknown functions with a minimum number of test points, has been applied to optimize the reaction conditions in continuous flow with a focus on lowering the costs and increasing both the productivity and the sustainability of the process, as evaluated through its Process Mass Intensity (PMI). Although this is a common chemical transformation, the fact that it happens in a biphasic medium with organometallic catalysis makes its optimization more challenging. Six variables were identified as important for the reaction system: residence time, temperature, equivalents of the nonlimiting reactant, mol % of catalyst, mol % of ligand, and solvent volumes. With less than 40 experiments, the BO algorithm achieved solutions that increased the productivity up to 4 times compared to the initial results of the DOE experiments and reduced the cost by around 30%. A Pareto front with different optimal solutions was provided at the end of the optimization, and the point with the lowest cost and the lowest PMI was scaled up to increase productivity while keeping the unitary cost unchanged.