Photo: Oak Ridge National Laboratory
February 3, 2021
The shift to remote work across the globe was one of the hallmarks of 2020. This shift has transcended industry and geography - and it has even impacted the scientists and researchers responsible for studying COVID-19, its effects, and potential treatments and vaccines. But even in this unusual working environment, a group of researchers across America went to remarkable lengths - with the help of IBM's Summit supercomputer - to better understand the structure and replication of the COVID-19 virus.
For their efforts, the group was recently awarded the first-of-its-kind Gordon Bell Special Prize for High Performance Computing-Based COVID-19 Research. The prize recognizes outstanding research achievement toward the understanding of the COVID-19 pandemic through the use of high-performance computing (HPC).
In their paper, the winning team developed an AI-driven workflow that leverages HPC to explore the time-dependent dynamics of molecular systems. They then used this workflow to investigate the mechanisms of infectivity of the SARS-CoV-2 spike protein, the main infection component of the COVID-19 virus.
We spoke to three of the scientists involved - Rommie Amaro, Professor and Endowed Chair, Chemistry and Biochemistry, UC San Diego; Arvind Ramanathan, computational biologist in the Data Science and Learning Division at Argonne National Laboratory; and Bronson Messer, Director of Science for the Oak Ridge Leadership Computing Facility at Oak Ridge National Laboratory - about their research, the impact of Summit and AI on their work, and how they were able to collaborate and succeed in a remote work environment.
Why did you use Summit for this research over other computing systems? How did it make a difference?
Messer: Summit is a unique platform, and I always say the machines we run have more in common with the Hubble Space Telescope or the Large Hadron Collider than the laptop or desktop in your home office. Summit allows for true scientific simulation rather than more common proofs of concept on other systems.
Amaro: Summit was made for this type of research. Our goal was to create a full, dynamic picture of the virus and the structure of the spike protein and how it interacts with other proteins, with human cells, and with antibodies.
Summit allowed us to look beyond static images of the spike protein in different isolated situations, and create a fuller, more realistic picture of the entire viral envelope, including the viral membranes and surrounding proteins. These simulations are also pretty big, and this one was among the most significant that has ever been successfully run. Using Summit was faster by a factor of 2-3x over other systems, which is considerable given the data we were working with.
How important were Summit's AI capabilities throughout the course of this research project?
Messer: Summit is the world's smartest supercomputer because of the depth and breadth of its AI stack and how it can be customized. That speaks directly to our mission and how we want to use Summit and make it available to people.
Ramanathan: I've called this a 'marriage made in heaven' because many of these AI models were customized for Summit to get the maximum performance out of it. There were certain operations and certain things that we optimized in such a way to get even more bandwidth. We were also doing the optimization of the settings of hyperparameters, which we had to get right as these simulations are running. So, there was an algorithmic side of the story where we developed a new model on the Summit system so we could train faster, and then we also looked at various ways in which we could get the optimized model to scale and learn from the data. The whole point was to take the entire virus simulation Rommie was running and ask the same questions.
Can you elaborate on the importance of your research for ongoing vaccine efforts, both for COVID-19 and future viruses?
Amaro: One of the main limitations of some experimental techniques is the inability to see the shield of sugars that surrounds cells in the human body. Viruses have evolved the ability to cloak themselves in this shield so that when they are in the body they don't get detected as an invader by our immune systems.
The COVID-19 vaccines in development were selected back in January, but with this computing we've been able to rebuild that sugary shield and better understand how it is moving. That was one of the main science outcomes here, learning more about what the spike protein really looks like. We wanted to know where the shield was, and just as important, where it was not, because that leaves areas of vulnerability. That is also where neutralizing antibodies can bind, and where drug designers could take aim with vaccines going forward.
You carried out this research in the same remote working environment that so many have experienced this year. How were you able to collaborate so successfully across different parts of the country on a project that was so important and time-critical?
Amaro: I have not been back to the office since March, so this was definitely a new way of working. These projects also typically require more formal proposals and evaluations that can be time consuming before the work begins. In this case, the project came out of the HPC Consortium, so we had access to Summit much faster. That was a special thing which allowed us to get to work more quickly. This is a very unique time to be a scientist and the project progressed very organically.
Ramanathan: There were 29 authors on the paper, so to have everyone working together and driving in the same direction was special. It was a newer venue for all of us to work in and feel part of a community. We were able to enable that through collaboration and it led to a lot of innovative thinking on the way we approached things. And of course, the Slack channel just exploded for all of us!