DEL Screening and ML for Hit ID for the WD40-Repeat Oncology Targets DCAF1, WDR5 and WDR12
Training machine-learned models using DNA-Encoded Chemical Library screening data “DEL-ML” and using these models to rank compounds in virtual catalogs has been successful for hit identification for a range of individual oncology targets including ER alpha and c-KIT [McCloskey et al 2020]. Here we describe the importance of library quality and diversity, protein quality, screen design, appropriate profile choice, training set preparation and evaluation, promiscuity labeling, chemical representations, modeling algorithms and their tuning in the context of a systematic target class-based evaluation for a range of oncology targets including DCAF1, WDR5 and WDR12. This approach yielded first-in-class, drug-like ligands for multiple targets that were characterized using a range of biophysical approaches including x-ray crystallography.