(MRC DTP) Applying machine learning to C-H functionalization for drug discovery

Drug discovery and development currently relies on a very small set of reactions that have traditionally been favoured over the past 20 years. As a consequence only a narrow section of the chemical shape space is being commonly explored during the great majority of drug discovery efforts. This low chemical structural diversity has been proposed as one of the causes of the large decrease in successful drugs to market over the last decades. In recent years, C-H functionalization has emerged as an extremely promising new synthetic tool that could accelerate drug discovery by allowing fast access to new chemical space through the direct modification of existing drugs or biologically active molecules at their C-skeleton structure. However, the adaptation of C-H functionalization tools to achieve this proposed ‘Late-Stage Functionalization’ faces major challenges in that most transition metal catalysts are incompatible with the highly polar functionalities present in biologically active molecules. The Larrosa group has recently developed a novel class of ruthenium-catalysts that are able to carry out C-H arylation on a wide range of drugs (see Nature Chemistry, doi: 10.1038/s41557-018-0062-3). Due to the complexity of these synthetic systems, however, the prediction of the position of C-H functionalization in the drug molecule cannot be fully achieved with current understanding, thus severely limiting the power for introducing novel functionalities in a controlled manner.

In a collaborative effort between the Larrosa group (School of Chemistry) and the Popelier group housed at the Manchester Institute of Biotechnology, this project will focus on applying state-of-the-art machine learning into the development of a predictive model capable of accurately predicting the position of C-H functionalization by a given combination of drug substrate, functionalizing reagent, and catalyst-ligand structure, and, conversely predicting the correct reagent and catalyst-ligand combination for C-H functionalization at a chosen position. The model created will provide a treasure of opportunities for controlled drug functionalization and derivatization, thus facilitating drug discovery and development. In particular, the machine learning features will benefit from the quantum character of the so-called topological atoms naturally emerging in heterocyclic rings.

The successful applicant will receive state-of-the-art training on all aspects of the project, including organometallic chemistry, catalysis, synthetic chemistry, generation of compound libraries for drug discovery (Larrosa group) and training in the systematic analysis of relevant molecular fragments (including novel heterocyclic rings) using the method of Quantum Chemical Topology in combination with Gaussian Processes initially (Popelier group).

http://personalpages.manchester.ac.uk/staff/igor.larrosa/index.html
http://www.qct.manchester.ac.uk