Using computation to design and explore compound libraries, with an emphasis on DELs

David Mobley, Chris Zhang, Ana Caldaruse

In recent work in the lab, we have been interested in using molecular modelling to help guide the design and screening of compound libraries, as well as subsequent experimental and computational follow up. In particular, our interests focus on collaborations with partners in the subset of combinatorial chemistry involving DNA encoded libraries (DELs), where chemical libraries are constructed via linking diverse building blocks using relatively straightforward chemistries, then following experimental screening, identities of active compounds can be read out via DNA bar code tags (which may or may not remain attached to product compounds). DEL data generates a wealth of activity data, and thus could provide training data for target-specific machine learning models, as well as for other types of models and additional computational and experimental follow-up. Here, we report on recent work from the group relating to DELs, including simple predictive models based on building block level similarities. We also report on work assisting design of experimental libraries for screening, and optimizing library cost while working to ensure sufficient coverage of chemical diversity. In addition, we are seeking to develop DEL-specific active learning workflows to efficiently employ multiple types of computation in addition with experiment to minimize time and costs spent in design cycles, so we report on early progress in that area as well. While many of the approaches discussed are relatively general, our particular focus is on computation for and with DELs.