Inside supercomputing's race to find coronavirus treatments and cures
Amazon Web Services says it has taken on 11 projects within the White House-led consortium, offering cloud computing to groups of researchers across continents.
At one of his daily press briefings in March, President Trump announced a unique project, a combined effort by government and private industry to accelerate COVID-19 research through the use of supercomputing.
Seven weeks later, the High Performance Computing Consortium, spearheaded by the White House's Office of Science and Technology Policy, the Department of Energy and IBM, is up and running, with 55 active projects and 95 research proposals submitted. The effort has brought longtime rivals including Amazon, Microsoft, Google and IBM together with dozens of researchers around the world, all working to fight the virus using complicated algorithms and enormous amounts of data.
Get what matters in tech, in your inbox every morning. Sign up for Source Code.
So far, the consortium has grown to 38 members and encapsulates 437 petaflops of computing power, said Paul Dabbar, the under secretary for science at the Department of Energy, in a webinar this week. It's a dramatic example of broad efforts by scientists and technologists to bring an array of skills to the pandemic response, providing software platforms and advanced hardware to aid and accelerate drug and vaccine studies.
"The White House applauds the important research being done by our partners in the COVID-19 HPC Consortium," U.S. Chief Technology Officer Michael Kratsios said in a statement Wednesday. "We look forward to continued progress."
The companies involved have offered few details about which projects they're supporting in particular and how their services are being used. But Amazon Web Services told Protocol this week it has taken on 11 projects so far, offering cloud computing resources to groups of researchers across multiple continents. "We saw an opportunity to bring the benefits of cloud and the things we've built, in particular the scale and elasticity pieces of what we do, to bear for the race for treatments, vaccine and research," said Dave Levy, vice president of federal business for AWS.
One group of 12 MIT researchers is using machine learning in AWS to design copies of a protein that could "sop up" the virus, said Kevin Esvelt, an assistant professor at the university. His team feeds vast amounts of data into a model that predicts which variants of a protein, sACE2, could help inhibit COVID-19. "We require all of that computing power in order to simulate this incredibly vast landscape of possible sACE2 proteins," he said, adding there was "no way" they could do it without the technology.
A group of four researchers at Children's National Hospital in Washington, D.C., is combining hundreds of data sets to identify genes that could potentially be targeted to treat COVID-19. Wei Li, a principal investigator and assistant professor focused on genetic medicine at the Children's National Research Institute, the hospital's academic arm, said initial analysis of 30 data sets helped them identify some genes that "appear to disrupt the virus infection."
AWS servers are also helping to power "Project Moonshot," an open-source drug discovery effort from a machine learning company called PostEra, which was launched six months ago by startup accelerator Y Combinator. "We're continuing to iterate to get to an antiviral as fast as possible," said Aaron Morris, PostEra's co-founder and CEO.
AWS allotted each research team a number of "credits" that allow them to use the cloud for free. Morris said his team speaks to dedicated AWS solution architects two or three times per week, parsing technical questions.
Overall, as of Monday, the consortium included at least 16 projects on the science of COVID-19, eight focused on how patients are handling it, and 26 on the cure, mitigation, diagnosis or treatment of the disease.
The initiative is run by a committee of industry representatives, government officials and scientific experts who process and approve applications from researchers, often within days. If an application is approved, the committee pairs the researchers with a company that can help.
"Think of it as speed-dating," Levy said. Beyond that point, he said, the White House hasn't gotten directly involved: "The consortium meets regularly and [members provide] progress updates, but there's no involvement in the work being done by the researchers."
Some officials have marveled that cutthroat tech rivals are working side-by-side. Microsoft and Amazon, for instance, remain in an ugly battle over a $10 billion Pentagon cloud computing contract. The effort involves "would-be competitors working alongside each other during this crisis to move the needle forward," Kratsios said.
Aside from AWS, it's unclear how many projects each company has helped launch. Google, Microsoft and IBM declined to provide figures.
However, Google said it is working on epidemiological modeling with Northeastern University researchers and applying deep learning to medical imaging with the Complutense University of Madrid. IBM said there are "a number of projects running on IBM and IBM-built systems that include such topics as finding new therapeutics and effective coronavirus drugs."
"The whole point of the consortium is to drive collaboration and support COVID-19 research efforts," an IBM spokesperson wrote in an email. "It's not a competition between system providers. The focus is simply to fight the virus."
The consortium is seeking more members as it plans to expand, particularly companies on the hardware side. "We'd like to see more industry participation in the process," said Jim Brase, deputy associate director at Lawrence Livermore National Laboratory. "That's one of the things we'd like to get help with."
Correction: The name of the supercomputing effort is the High Performance Computing Consortium. This story was updated on May 21, 2020.