Baptiste Canault began his presentation by expressing gratitude to the conference organisers for the opportunity to discuss annotating chemical space for automated drug design. He introduced the BRADSHAW system, an automated platform developed by GSK to optimise the drug discovery pipeline. The primary aim of BRADSHAW is to reduce the time required to identify or select candidate compounds.
Canault explained the molecular design process, which involves designing experiments, synthesising molecules, testing them biologically, and analysing the results. This iterative cycle aims to quickly identify candidate compounds by continuously reviewing project expectations. The BRADSHAW system uses various molecule generators, including reaction-based, knowledge-based, and deep learning-based generators, to produce thousands or millions of chemical ideations. The challenge was to filter this vast amount of data to generate a manageable list of compounds for further processing by chemists.
Canault highlighted the importance of chemical annotation and introduced the CAScore, a new chemical dissimilarity matrix developed by GSK. CAScore is designed to make distinctions between small modifications around substitutions and scaffold modifications, providing a better estimation of the rank of compounds. The CAScore uses machine learning, specifically random forest models, to train and validate the score. This approach helps in extracting knowledge from datasets and setting the CAScore accurately.
Canault compared CAScore with traditional approaches, demonstrating its efficiency in screening a smaller percentage of the ChEMBL database to recover a significant portion of Factor Xa molecules. He emphasised that CAScore could prioritise compounds more effectively, reaching a better granularity of chemical ideation. The presentation concluded with Canault summarising the benefits of CAScore, including its ability to provide a good quality ranking and prioritise compounds of interest.
