Addiction to opioid drugs causes nearly 50,000 deaths a year in the United States—and has become a major public health crisis. One solution is pharmacological: to develop an alternative to opioid medications that can relieve pain without the high risks of addiction and fatal overdoses. “That’s our goal,” says Vsevolod Katritch, Ph.D., Assistant Professor of Biological Sciences and Chemistry at the University of Southern California (USC). As an affiliated faculty member with USC’s interdisciplinary Michelson Center for Convergence Biosciences and the Center for Drug Discovery and Development, Katritch runs a lab that uses bioinformatics and molecular modeling to understand the structure and function of G protein-coupled receptors, which mediate many of the human body’s responses to external stimuli. Nilkanth Patel, a Ph.D. candidate in Katritch’s lab, explains that “you can think of opioid drugs as keys, which can selectively open or block proteins in our brain called opioid receptors. Basically, we want to find and design better keys—in this case opening the receptor that relieves pain, while blocking another that is associated with addiction.” By searching for other chemical compounds that can link to those receptors, Katritch and his team hope to discover a new generation of drugs that can become a safer alternative to opiods.

Developing new drugs is a complex, time-consuming process that can take as long as ten years and cost millions of dollars. Before pharmaceutical companies can even begin trials, research chemists need to find viable drug candidates, or ligands that bind to a specific receptor protein. Katritch’s lab uses Virtual Ligand Screening to test a vast library of possible chemical compounds by modelling each one through thousands of possible orientations into the binding pocket to find a matching protein receptor—like fitting pieces of a 3D jigsaw puzzle together. “Computation and computer-assisted design on a large scale are changing the paradigm for drug discovery,” Katritch says. His library contains over half a billion virtual compounds now and he expects that number to grow to ten billion in the near future. To analyze this vast amount of data, Katritch’s lab turned to Onix, a leading cloud services solutions provider, and Google Cloud .

Cloud computing speeds discovery of candidate drugs from one year to one day

With funding from a National Institute of Health (NIH) pilot program to encourage cloud computing for advanced biomedical research, Onix and Katritch’s lab designed a workflow to use the power and speed of cloud computing and the accuracy of Virtual Ligand Screening software from Molsoft LLC to make the search for candidate drugs more efficient. They divided the huge job into small batches and ran them over weekends using Google’s pre-emptible VM instances on 200K CPU cores. After several iterations, the team narrowed down 680 million possible compounds to about 120 promising candidates for further testing. “We found using Google Compute Engine VMs really convenient, flexible, and cost-effective,” Katritch says. “Doing this job on our lab cluster of 600 cores would have taken one year. Instead we did it in 24 hours on Google Cloud. And with Onix’s help we completed the job on budget as well as on time.”

Cloud computing revolutionizes drug discovery

Katritch’s lab has been working on possible drug alternatives for chronic pain for several years and discovered some potential leads. In the meantime, his team continues to double check the initial large-scale screening results from Google Cloud to synthesize even more promising candidate leads for testing at Dr. Bryan Roth’s lab at the University of North Carolina’s Department of Pharmacology. For Katritch, the process already proves the value of the NIH pilot program and the benefits of cloud computing in improving the drug discovery process: “you never find the ideal drug immediately. You find the hit, which is pharma slang for narrowing your choices. From identifying a hit to finding a lead to testing a drug candidate there are many steps and many parameters. So when we narrow the hits in this huge ocean of diverse and similar compounds it makes the path to from a hit to a drug candidate much easier, faster, and also more efficient. Cloud computing streamlines the whole process. This type of large-scale CPU and GPU computation is really a game changer for us.”