The grant is funded by AstraZeneca and has been awarded to A Gammerman (PI), V Vovk (CI) and Z Luo (CI).

An important task in pharmaceutical industry is to predict bioactivity of chemical compounds. Success in solving this task would allow to select potentially successful drugs. It is known that each chemical compound interacts with reagents and produce either a new compound or a modified old one.

The project aims to select machine learning strategies to control which chemical compounds to synthesise and in what order to test them. Given a huge search space the task of reducing empirical testing and the associated cost is a difficult problem. An objective of the research is to investigate a possible way to reduce the number of tests needed to identify a compound as Candidate Drug at the Lead Optimization stage. This can be done by discarding compounds that we predict as having a negligible probability to being eventually selected as Candidate Drug.

The project will consider recently developed methods of Conformal Predictive distributions and their application to Decision Making. The aim is to find an effective way to synthesise the new compound with some desirable properties by forming a suitable objective function subject to certain constraints. The project will use publicly available data as well as data from the company.