Flowing with intelligence Machine learning driven screening, optimization and prediction of photocatalytic reactions in flow

This thesis presents the development of intelligent, machine learning-guided workflows for the optimisation, screening, and prediction of chemical reactions. The work bridges the fields of chemistry, automation, and data science, with the aim of improving how experimental data is collected and used to drive discovery in synthetic chemistry.
The thesis spans three main projects. The first focuses on RoboChem, an automated flow platform designed for self-optimisation and scale-up of photocatalytic reactions. By integrating Bayesian optimisation, RoboChem demonstrated efficient enhancement of reaction yield and throughput in a reproducible and data-rich manner.
The second project explores a high-throughput experimentation workflow for reaction screening. Using an initial dataset and a Random Forest model, the method predicts promising reaction conditions in unexplored chemical space. This approach successfully identified new C(sp³)–C(sp³) bond-forming reactions.
The third project centres on reaction prediction, particularly using data collected via RoboChem. Substrates were clustered to guide targeted experimentation, generating a dataset that enabled reaction outcome predictions under various conditions. While further data is needed for broader generalisation, the results show the predictive potential of RoboChem-generated data.