DNA-encoded small molecule libraries (DELs) have enabled discovery of novel inhibitors to many distinct protein targets of therapeutic value through screening of libraries with up to billions of unique small molecules. Building on this technique, we demonstrate a new approach to hit finding by training machine learning models from DEL selection data and applying these models to identify active compounds from both a large commercial collection and a virtual library of easily synthesized compounds. We validate this approach with a large prospective study of three diverse protein targets (a hydrolase, a nuclear receptor, and a kinase) with nearly 1900 compounds tested. The approach is effective, with an overall hit rate of ∼30% at 30μM and discovery of potent compounds (IC50&lt;10 nM) for every target. The model makes useful predictions even for molecules dissimilar to the original DEL, and the compounds identified are diverse, drug-like and different from known ligands. Collectively, the low marginal costs of DEL selections, the power of modern machine learning methods, and access to large, inexpensive, commercially-available libraries opens up a powerful new approach to hit finding.

Machine learning on DNA-encoded libraries: A new paradigm for hit-finding

ACS Fall 2020 Virtual Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.