Highly Parallel Optimisation of Nickel-Catalysed Suzuki Reactions through Automation and Machine Intelligence

We report the development and application of a scalable machine learning optimisation framework for batched multi-objective reaction optimisation. Through experimental data-derived benchmarks, we demonstrate our approach’s capacity to efficiently handle large parallel batches and high-dimensional search spaces characteristic of high-throughput experimentation (HTE). We also establish the framework’s robustness to reaction noise and handling of batch constraints encountered in real-world chemical laboratories. We applied our approach experimentally through an automated 96-well HTE reaction optimisation campaign for a nickel-catalysed Suzuki reaction, aiming to tackle challenges in non-precious metal catalysis. Our optimisation framework effectively navigates the complex reaction landscape with unexpected chemical reactivity, revealing advantages over traditional, purely experimentalist-driven HTE plate design. By integrating machine intelligence with highly parallel reaction execution via HTE robots, this work aims to accelerate reaction optimisation in academia and the pharmaceutical industry. This workflow can also be extended beyond HTE settings to any chemical reaction of interest.