Automated High Throughput Machine Learning-guided Condition Optimisation for Biocatalysis

Making a pharmaceutical drug often fails not because the chemistry is impossible, but because finding the perfect conditions—temperature, pH, concentration—takes months of trial and error, especially when using enzymes as catalysts. This project tackles that bottleneck. Enzymes (biocatalysts) are the most sustainable way to run chemical reactions, but they are also the most complex to optimise because their behaviour depends on thousands of variables. Current machine-learning approaches have streamlined optimisation for conventional chemical reactions, but they have not yet been successfully applied to biocatalysis. The researchers will combine automated high-throughput experimentation with machine learning, and crucially, feed in molecular dynamics simulations—computer models of how enzymes move and flex at the atomic level—to guide the optimisation process. If it works, the impact is on pharmaceutical manufacturing. A few percentage points improvement in reaction yield can determine whether a new drug is economically viable to produce. More broadly, making biocatalysis faster and cheaper to develop would reduce the carbon footprint of chemical manufacturing and enable circular-economy processes that recycle waste into valuable products—systems most people never see, but which quietly underpin modern medicine and materials.