Machine learning approaches to reaction design and optimisation

The design and synthesis of tailored molecules is of high importance in the pharmaceutical industry. Frequently, trialand-
error methods are used to design these molecules however, this can become time consuming with a high
demand on resources. Research groups and companies are searching for quicker and more efficient methods of
bespoke molecule design that are more sustainable.
Computational design has been used both within industry and academia to good effect throughout the years, with
quantum mechanical calculations playing a large role. These calculations can become time consuming when the
systems become more complicated - especially when compared to traditional experimental screening methods (highthroughput
experimentation). Exploring the conformations of catalysts and substrates using quantum mechanical
calculations is time consuming. To be able to optimise reactions without the need for multiple experimental screens
will allow chemists across various industries to focus further on chemistry and less time on experimental design. This
will also greatly improve the sustainability surrounding reaction design and optimisation.
This project, funded by AstraZeneca, will produce machine learning models that can rapidly predict the outcomes of
catalytic reactions thus replacing the need for time-consuming quantum mechanical calculations and extensive
experimental design.