Abstract
Screening diverse chemical structure compounds is an essential task in modern drug discovery. It provides different opportunities to avoid patent invasion, avoid potential toxicity observed in similar compounds, and execute new potential pharmacological functions. G protein-coupled receptors (GPCRs) span an important family of membrane proteins that play a central role in signal transduction and serve as significant drug targets. A prototypical class A GPCR is the β(2)-adrenergic receptor (β(2)AR), which is widely targeted by agonists to treat respiratory diseases. Although various β(2)AR agonists are currently available on the market, there is still an urgent demand for further optimizing drug safety, efficacy, and selectivity. Here, we combine machine learning (ML) methods with other computational methods to efficiently screen agonists against β(2)AR from a large compound library, composited of 19 million molecules. Verified by cellular functional assays, we identified several extremely potent agonists showing EC(50) values in the range of 0.2-20 nM with new chemical structures, of which the structure is diverse from previous reported molecules. Our machine learning computational approaches provide new possibilities to design novel drug candidates for GPCR.