Prior knowledge-based multi-round multi-objective Bayesian optimization: continuous flow synthesis and scale-up of O-methylisourea

Optimizing and scaling up chemical reactions is a critical step in transitioning from laboratory to industrial production, often involving trade-offs between multiple objectives such as production efficiency and cost. Meanwhile, the traditional synthesis process for the key pharmaceutical intermediate O-methylisourea no longer meets the demands of modern smart and green production, requiring urgent improvement. This study demonstrates the use of multi-objective Bayesian optimization (MOBO) for the continuous flow synthesis of O-methylisourea. By leveraging historical reaction data, we accelerated optimization as parameter ranges changed, enabling rapid inference and adjustment. Using a scaled-up continuous flow system equipped with MOBO and transfer learning capabilities, we identified optimal conditions along the Pareto front, achieving a production rate of up to 52.2 g/h and an E-factor as low as 0.557 during the third round of optimization, while maintaining a yield of approximately 75 %. These results highlight the scalability and efficiency of Bayesian optimization in accelerating reaction optimization and facilitating the transition from laboratory to industrial-scale production.