A Bayesian machine learning approach for drug target identification using diverse data types

It typically takes 15 years and 2.6 billion dollars to go from a small molecule in the lab to an approved drug1,2,3, and for natural products and phenotypic screen derived small molecules, one of the greatest bottlenecks is identifying the targets of any candidate molecules2,4. Proper understanding of binding targets can position drugs for ideal indications and patients, allow for better analog design, and explain observed adverse events. There exist a number of experimental approaches for target identification ranging from affinity pull-downs to genome-wide knockdown screens4,5, but these approaches are labor, resource, and time intensive, not to mention failure prone. Computational approaches have the potential to substantially reduce the work and resources needed for drug target identification. Traditionally, ligand-based approaches take known binding targets for a given drug and attempt to find other drugs or proteins that are sufficiently similar6.