Data-driven reactivity prediction using computed quantum features for drug discovery

Sector: Pharmaceuticals & Drug Design

Lead organisation: Capgemini UK Plc

Consortia: Capgemini UK Plc, GSK, National Quantum Computing Centre (NQCC)

Within the pharmaceutical and drug design sector, this project, Capgemini UK Plc, in collaboration with GSK and the NQCC, tackled a central challenge in the design of targeted covalent drugs which lies in the reactivity of the ‘warhead’. The partners on this project had already developed an approach to warhead reactivity prediction, which centres on deriving chemical features from quantum calculations. Crucially, this project aimed to find a general approach where rich quantum-derived features can be used for downstream modelling without a significant understanding of reaction mechanisms being known a priori. To date, work has almost entirely been performed on quantum simulators. The focus of this project under the POC call was to identify how to calculate such quantum features so that they are both informative for predictions such as reactivity and can further be calculated on quantum hardware in a scalable way.

Focussing on predicting the reactivity of a chemical series of sulfonyl fluoride compounds, a robust software pipeline was created, which allowed data-driven workflow using quantum features of molecules to make predictions using a machine learning model. The approach potentially offers better generalisation from fewer measured examples, compared to conventional machine learning techniques. The project concluded that the quantum-driven reactivity prediction approach has the potential to significantly accelerate the molecular design process, and eventually, reduce costs and allow for higher throughput.