Rxn Rover: automation of chemical reactions with user-friendly, modular software†
Abstract
The automation of chemical reactions in research and development can be an enabling technology to reduce cost and waste generation in light of technology transformation towards renewable feedstocks and energy in chemical industry. Automation of reaction optimization, in particular, would remove the need for expert input by designing algorithms to statistically analyze the reaction and automatically generate suggested results. In addition, automation can save time and resources, and reduce random human error. However, automation software is commonly coupled to a specific laboratory or device setup or not freely available for use. Rxn Rover is an open-source, modular automation platform for reaction discovery and optimization. Primarily targetting smaller research groups, it is designed using interchangeable plugins to be flexible and easy to integrate into a variety of laboratory environments. Using the Rxn Rover plugin architecture, novel optimization algorithms, analysis instrumentation, and reactor components can be used with minimal or no programming experience. The capability of Rxn Rover is demonstrated in the optimization of a reduction reaction of imine to amine, relevant to energy conversion and manufacturing of fine and commodity chemicals. The reaction was optimized separately using optimizer plugins for SQSnobFit, a Python implementation of the SNOBFIT global optimization algorithm, and Deep Reaction Optimizer (DRO), a deep reinforcement learning algorithm designed for reaction optimization. Using plugins designed for pumps, temperature controllers, and an online liquid chromatography system, the flow reaction was able to be controlled by each algorithm to automate reaction optimization for up to three days, at which point the results were gathered. A successful optimization was performed with SQSnobFit, achieving 70% yield and 95% selectivity, while no successful optimizations were achieved with DRO. Regardless of algorithm performance, Rxn Rover was able to successfully automate both multi-day optimization searches.