(259a) Data Rich Experimentation and Machine-Assisted Process Development for Rapid Reaction Optimization

Identifying the optimal conditions for complex reactions is a challenging task in drug substance development and can be hindered by material availability, program timelines, limited data, and experimental bias. However, the combination of parallel reactor technology with automated sampling strategies and machine learning approaches can greatly accelerate the optimization procedure with minimal user intervention. To demonstrate these improvements, we applied these technologies to optimize two case study reactions – the copper catalyzed methoxylation of an aryl halide and the multihalogenation of an oligosaccharide. In each case study, reactions were conducted in a reactor block that enabled the execution and sampling of 10 reactions simultaneously. Data from these reaction profiles were inputted into a direct search optimization algorithm, which provided the conditions for the next round of 10 experiments. In this manner, the reactions were optimized in one week using 40 experiments and a minimal amount of material. Data analytics and kinetic modeling provided mechanistic understanding and further process insights using the same data set obtained from the optimization procedure.