Meet your new lab assistant: an automated flow chemistry system called RoboChem that uses machine learning to optimize a range of light-driven reactions (Science 2024, DOI: 10.1126/science.adj1817). It can tweak conditions such as the concentration of reagents, loading of catalyst, and the intensity of light to improve the yield and throughput of reactions including carbon-hydrogen alkylation and C–C coupling.
This kind of optimization can be particularly time consuming for photocatalytic reactions. For example, they can be highly sensitive to the intensity of light, and ideal reaction conditions may vary significantly between similar substrates or with different lab equipment.
Rather than developing generic reactions conditions that could be applied across a whole class of substrates, “we wanted an individual optimization, so that you have tailored reaction conditions for every single substrate,” says Timothy Noël, who led the University of Amsterdam team behind RoboChem.
The system can optimize the synthesis of about 10–20 different molecules per week, a much faster pace than a human chemist.
The team tells RoboChem which reaction conditions it should vary within a given range. Then, the system runs experiments in 650 µl slugs of liquid that flow through tubes to a photoreactor illuminated by light-emitting diodes. After analyzing the products using a benchtop nuclear magnetic resonance spectrometer, RoboChem’s machine learning model plans the next iteration of the reaction to boost performance.
In some cases, the process discovered conditions that doubled the yield of product compared with published syntheses. The team now hopes to adapt RoboChem so that it can optimize reactions driven by heat or electricity, for instance.