Beyond the microbiological and pharmacological work of examining how SARS-CoV-2 operates and identifying compounds that might inhibit it, there is the epidemiological work of examining how COVID-19 spreads and identifying what policies might inhibit it. For the University of Texas at Austin’s COVID-19 Modeling Consortium, that’s their bread and butter: throughout the year, its researchers have been applying computer modeling to investigate how various policy measures could slow the pandemic. A recent feature from the Texas Advanced Computing Center (TACC), which supported the research, charted how the modeling has evolved over the last nine months.
Early in the pandemic, the consortium released a model projecting mortality across the U.S. by examining case numbers in combination with estimates of how closely people in any individual metropolitan area were interacting with each other. “The model uses geolocation data from cell phones to determine the impact of social distancing within each specific place,” explained Lauren Ancel Meyers, who leads the consortium . “Using more granular cell phone mobility data, UT can project COVID-19 deaths for the next three weeks.”
These projections are made possible by supercomputers – and in particular, by Frontera. The Frontera system, hosted by the Texas Advanced Computing Center (TACC), is equipped with 8,008 Intel Xeon-powered compute nodes (each with 192 GB of memory) and Nvidia/Mellanox InfiniBand HDR100 networking. Frontera also has two Nvidia GPU-powered subsystems. Overall, Frontera delivers 23.5 Linpack petaflops, placing it ninth on the most recent Top500 list.
Over time, the modeling evolved for this urgent, high-intensity use case. Computational experts at TACC updated code to make it run smoothly on TACC’s supercomputers, enabling virtually real-time analyses early in the pandemic.
But it’s not just the technical modeling that sets UT Austin’s efforts apart – it’s the communication with the general public. From the beginning, the consortium put a strong emphasis on reaching individuals to help inform their decision-making processes.
If you visit the dashboard today, you’ll be greeted by an interface that lets you examine mortality projections for any major metropolitan area in the country; projections for hospitalizations, the COVID-19 reproduction rate and ICU patients in any city in Texas; and even more projections specifically for the Austin area. The consortium has even introduced a school risk dashboard for the U.S., where users can estimate the risk for a school of a given size anywhere in the country.
“What makes this situation special is the risk communication aspect of it,” said Kelly Gaither, director of Health Informatics at TACC. “It’s high stakes. We’re not just communicating to a scientific audience. The audience for this information extends from the lay public to the researcher to decision-makers, so we have to communicate in a range of ways.”
“At the end of the day, models are impactful,” added Maytal Dahan, director of Advanced Computing Interfaces at TACC. “By expanding the model with visualization and interactive web access via a science gateway, we’re able to make the most impact for public understanding.”
UT Austin and TACC are pleased with how the models have been received and utilized by the general public and policy-makers.
“The mortality projections website that the team at TACC put together has helped our COVID-19 modeling work have a public-health impact beyond anything we could have hoped to achieve ourselves,” said James Scott, one of the model’s creators and an associate professor of Statistics at UT Austin. “Because of the visibility and user-friendliness the site offers, our projections have garnered nationwide attention from Anderson Cooper to the CDC, and from the Rhode Island Governor’s Office to the Department of Defense.
Still, with cases continuing to skyrocket, the consortium has plenty of plans to update the model to increase its utility even further, including comparative visualizations and a bigger emphasis on specific geographies.
“Right now, we’re citing risks to states, cities, counties, or metro areas,” Gaither said. “But we hope to get to the point where we can analyze what’s going on at the school district, zip code or neighborhood level, like having a digital twin of a community.”
To read more about UT Austin’s modeling, visit Aaron Dubrow’s article here.