Outside the U.S. National Science Foundation headquarters in Alexandria, Virginia, an unexpected light snow was falling, adding to a feeling of excitement for Madhav Marathe and his team of researchers from the University of Virginia and across the world.
Inside, in a conference room, Marathe queued up the first slide in a PowerPoint presentation the team was giving to a group of more than a dozen NSF directors and review committee members from the Directorate for Computer and Information Science and Engineering.
A pandemic will occur: The question is not IF but when.
The slide’s bold statement was prescient.
Marathe, professor of computer science at UVA’s School of Engineering and Applied Science and distinguished professor at UVA’s Biocomplexity Institute, and the team of multi-disciplinary scientists representing 14 U.S. institutions and 20 international organizations were focused on developing new computational tools to help policy makers, health care providers and everyday citizens handle global epidemics.
The presentation of their project on this snowy day was the final phase of a very long journey to win a coveted NSF Expeditions in Computing grant – a highly competitive, five-year, $10 million award aimed at propelling science in the quest to answer big societal questions.
Since the early 1900s, most of the framework for modeling how infectious diseases spread was based on differential equations developed by 1902 Nobel Prize winner Ronald Ross, a British scientist who used math to study malaria. Ross’ method came out of the worlds of physics and chemistry, where molecules interact with each other and the reaction creates new molecules. In some ways, scientists believed, this simulated the way humans interact and possibly spread disease.
That early approach provided basic estimates and only required the use of a pencil and paper. However, it didn’t, and couldn’t, account for the complexities of human-to-human and human-to-animal interactions and the detailed interactions people have every day, referred to by today’s scientists as multi-scale, multi-layer networks. These interactions have grown by orders of magnitude since Ross worked out his first equation, as populations have increased exponentially, global travel has become routine, and diseases have emerged that are resistant to treatments.
Today, Marathe and other computer science engineers and scientists weave together high-performance computing, machine learning, and artificial intelligence to run complex simulations that incorporate many different kinds of datasets, like activity data that show how people move over time, fitness-tracking data, health-related data, weather data and census data, to name a few.
“Our hypothesis was that the structural features of the social contact network, i.e., patterns of interactions between individuals [referred to as nodes] in a social network, have a tremendous impact on the outcome of epidemics and that they should be represented as faithfully as they can,” Marathe said. “This had become a computer science question, and now computer science was mature enough to undertake it.”
In the years prior to their NSF presentation, Marathe and his colleagues at the Biocomplexity Institute had many successes using computer science to model the spread of infectious diseases, but there were still a lot of questions. They had much research to do in applying computational modeling to real-time epidemiology – research they hope will one day give people more reliable information than ever before to develop intervention plans before infectious diseases reach pandemic proportions.
Sitting at the conference table on that wintry, December day, Marathe knew they had assembled a brilliant cohort of researchers eager to earn the NSF’s support for their very important, complex work, and that the chances were pretty good they might get the grant.
Still, there was one critical thing that was totally out of their control, although fundamental to the project: To do real-time epidemiology, they would need to study an infectious disease somewhere in the world during the five-year grant.
Infectious diseases cause more than 13 million deaths per year worldwide. Rapid growth in the human population and its ability to adapt to a variety of environmental conditions has resulted in unprecedented levels of interaction between humans and other species. This rise in interaction combined with emerging trends in globalization, anti-microbial resistance, urbanization, climate change, and ecological pressures has increased the risk of a global pandemic.
Computation and data sciences can capture the complexities underlying these disease determinants and revolutionize real-time epidemiology – leading to fundamentally new ways to reduce the global burden of infectious diseases that has plagued humanity for thousands of years.
- Excerpt from the abstract submitted to the NSF by Marathe and the research team, written in 2019
As the program director for the NSF’s Expeditions in Computing grant since 2009, Mitra Basu has read and evaluated hundreds of well-conceived proposals spanning a range of computer science disciplines focused on huge societal problems. In that time, however, her group has awarded only about 25 grants. The NSF Directorate for Computer and Information Science and Engineering established the Expeditions in Computing program in 2008 to provide funding to higher education and research institutions – giving them financial muscle to embark on ambitious, fundamental research initiatives that could dramatically advance the future of computing and information while building and strengthening multi-disciplinary collaborations necessary to tackle massive problems.
“The first thing the grant does is to advance the science,” Basu said. “Second is to train and prepare students, and the third area of focus is to see that research outcomes have great societal and economic benefits.”
Research teams spend years building up the expertise needed to earn an expeditions grant.
The basis of Marathe’s team’s proposal to NSF can be traced back two decades to the Los Alamos National Laboratory. There, Marathe worked with colleagues Christopher L. Barrett, Stephen Eubank and Anil Vullikanti devising a novel way to study an epidemic science problem. Part of this work was funded through the National Institutes of Health Models of Infectious Disease Agent Study project for 13 years. Barrett is now the UVA Biocomplexity Institute’s executive director, distinguished professor of biocomplexity and a professor of computer science at UVA Engineering. Vullikanti is a professor in the Network Systems Science and Advanced Computing Division of the Biocomplexity Institute and professor in UVA Engineering’s Department of Computer Science. Eubank is the institute’s deputy director in the Network Systems Science and Advanced Computing Division and a professor in the UVA School of Medicine’s Department of Public Health Sciences.
Initially, the group developed models using high-performance computers to study infrastructure, transportation and communications problems for the U.S. Department of Transportation, Department of Homeland Security, Department of Defense and other governmental organizations, and they wanted to apply their research to human health, specifically infectious disease spread.
Over the years, the group began actively working with various governmental organizations to deal with a variety of infectious diseases cropping up around the world. The team took an active role working with the NIH, the Defense Threat Reduction Agency and the Centers for Disease Control and Prevention to model outcomes for the H1N1 Swine Flu in 2009, MERS in 2012, Ebola in 2014 and Zika in 2016.
The depth and precision of their modeling had evolved as datasets grew in number, scale and type; high-performance computing became more available and powerful; and machine learning advanced. Marathe’s team and list of collaborators had also grown to include highly successful colleagues from around the globe working in diverse disciplines like ecology and evolutionary biology, microbiology, economics and social decision analytics.
At the end of December 2019, after months of poring over proposals and conducting site visits, Basu and her team concluded that Marathe’s team’s proposal, despite the caveat of needing an actual epidemic to study, checked all of their boxes and was worthy of NSF expeditions funding. Surely there would be some minor flare-up somewhere in the world in the next five years where they could apply their research.
“I wish I could claim that I could see the future,” Basu said. “At the time, we had no idea that such a black cloud and devastating storm was waiting beyond the horizon when we made the decisions.”
Just as Basu and her team were preparing to tell the principal investigators of their grant decisions in early January 2020, the World Health Organization, the CDC and the Chinese government were quietly sharing information about an outbreak of a pneumonia-like virus that had sprung up in Wuhan, China.
By mid-January, the first confirmed cases of the novel coronavirus, SARS-CoV-2, later known as COVID-19, had shown up in countries outside of China while TV news programs showed the Chinese rapidly building a pair of massive 2,000-bed field hospitals.
On January 22, 2020, the WHO confirmed human-to-human spread of the disease, and then-president Donald Trump went on national TV to proclaim, “We have it totally under control.”
Right about that time, Marathe got the word from Basu: His team had earned the NSF Expeditions in Computing grant, and the news would be publicly announced on March 24, 2020.
“There is no better way to drive science than with some urgency,” Basu said.
A year after Marathe shared his prognostication of an impending pandemic during his pitch to NSF officials, more than 330,000 people in the United States had died of COVID-19 and the virus was still raging coast to coast with record death tolls spiraling upward in December 2020.
The UVA-led team had wasted no time while awaiting the grant announcement.
Building on funds his group received from the Department of Defense, the NIH, the CDC and the Virginia Department of Health, and adding to the group’s past collaborative experience in dealing with the H1N1 and Ebola outbreaks, they began providing weekly briefings to state and federal officials, as well as members of the UVA COVID-19 task force almost immediately when the virus picked up steam in the United States and the country began shutting down.
The funding allowed the group to work 24 hours a day. “We are going on more than 100 weeks of continuous work,” Marathe said. "While we expected we'd have a virus to study during our five-year grant, even we didn't imagine that we'd be facing a global pandemic lasting more than two years.”
From UVA, the team includes Marathe; co-principal investigator Vullikanti; Barrett; Eubank; and Biocomplexity Institute researchers Bryan Lewis, S.S. Ravi, Daniel J. Rosenkrantz, Richard E. Stearns and Samarth Swarup.
Beyond UVA, the expeditions team includes a diverse group of scientists and researchers with more than 600 combined years of experience in related fields. UVA is joined by 12 institutions, including: Arizona State University; the Center for Disease Dynamics, Economics & Policy; Indiana University; Lawrence Livermore National Laboratory; Massachusetts Institute of Technology; Oak Ridge National Laboratory; Princeton University; Stanford University; State University of New York at Albany; University of Maryland; Virginia Tech; and Yale University.
“When you engage in these kinds of interdisciplinary interactions, where everybody comes to the table with a specific expertise, you can really advance knowledge quickly,” said Simon Levin, the James S. McDonnell Distinguished University Professor in Ecology and Evolutionary Biology at Princeton University, winner of the National Medal of Science and the director of the Center for BioComplexity at High Meadows Environmental Institute.
On December 15, 2020, the group publicly announced the creation of the COVID-19 Medical Resource Demand Dashboard, led by Mandy Wilson, research scientist at the Biocomplexity Institute. The dashboard could give information on hospitalizations and available hospital beds up to six weeks in advance. The dashboard has been a boon for health care administrators and state health agencies.
This dashboard was the only one of its kind to combine multiple forecasting models to project future hospital occupancies, allowing users to adjust projections based on multiple factors, such as the percentage of non-COVID hospital patients and the average duration of patient stays.
The team also brought to life real-time epidemiology through high-performance computing and multi-layer, multi-scale modeling. They forecasted for policy makers how such activities as contact tracing initiatives, lockdowns and mask mandates could lessen the disease’s spread in Virginia. The result of the team’s work was evident in the statewide data as policy makers’ decisions were playing out – Virginians largely followed the science brought to them by the team.
“In a state where we are not politically homogeneous, we fared pretty well,” Marathe said. “We owe this to the collective efforts of the state officials, business leaders, health care professionals and the citizens. And I think the results are there to see – our state fared better than most other states in terms of the overall public health impact.”
Not since the Spanish flu of 1918 had the world experienced a pandemic of the proportions on display with COVID-19, and there were many new revelations that affected how the group’s models were constructed.
“The political discourse around pandemics is huge. The role of misinformation is huge,” Marathe said. “It's becoming clearer to everyone that this is really not just a biological science problem, but in fact largely a problem of social science.
“In our early work, we were one of the first to say that pandemics don't live in isolation, they co-evolve with individual behavior, public policies, and the networks themselves,” he said. “This statement became absolutely central to the study of COVID because you could see the ebbs and flows of COVID during the course of the year, which were all due to changes in individual behaviors, like not wearing masks, and vaccine hesitancy. And adding these layers of representation to our modeling was crucial to forecasting outcomes.”
In the short but relentless period leading up to December 2020, the expeditions team not only conducted policy briefings, ran data simulations and built new computational tools, they also combined to publish more than 30 academic papers, which appeared in the journals Science, Nature and Proceedings of the National Academy of Science. Team members gave more than two dozen presentations all over the globe on topics ranging from predicting and analyzing pandemics to recommendations for improving science during crises.
“This pandemic has opened a number of areas of research, and many data and computing resources have now become available,” Vullikanti said. “Using them effectively for future pandemics will be a research focus.”
Most important, however, the team’s work literally saved lives. Virginia’s rate of infection per 100,000 people is among the lowest in the country, according to CDC data.
“The UVA Biocomplexity Institute has been an essential part of Virginia's response to the COVID-19 pandemic,” said Justin Crow, director for the Division of Social Epidemiology at the Virginia Department of Health. “In the earliest days of the pandemic, the modeling showed that containment measures were buying us time, and we were able to pause rolling out field hospitals, diverting those resources to other efforts. They were able to anticipate last winter's surge and the Delta wave, giving us several weeks to prepare for these. Outside of the projections, they've helped to justify and shape mitigation, test and trace, and genomic surveillance efforts. The team has been responsive, innovative, and often able to anticipate needs.”
The Biocomplexity Institute also played an essential public education role, Crow said. “In addition to our programs, health systems, colleges and universities, K-12 schools, local governments, businesses and the press all used the weekly updates to guide decisions. There is no doubt the UVA team's work gave Virginians the information they needed to make good decisions and had a large impact on the course of the pandemic in Virginia.”
The computational tools the team developed allowed decision-makers to see aspects of the virus’ trajectory that wouldn’t have been visible before, Marathe said.
“Our work has been invaluable because it has changed the narrative,” he said. “It has allowed us to come up with new interventions that we likely would not have thought of. Doing science in real-time like this, which we anticipated, but didn’t anticipate it going on for this long or with this level of intensity, has fundamentally changed people’s viewpoints.”
A Look back at 2021
In more than two years since that snowy visit to NSF headquarters, things haven’t slowed down for the team in the slightest.
As 2021 began, less than 5% of the U.S. population had received a vaccine, and the expeditions team added modeling for the vaccine rollout to the quiver of data they were sharing with state and federal officials on a near-daily basis.
A Nature article about how human mobility patterns have changed during the pandemic and how that has affected virus spread, authored by expeditions team member Jure Leskovec, associate professor of computer science at Stanford, was cited during two U.S. Supreme Court decisions to uphold state-level vaccine mandates.
Infection numbers generally fell around the country through the spring and summer of 2021, save a few hot spots like Florida and Texas. Some places in Virginia didn’t report a single case of COVID-19 for days. But the forecasting trend was about to change. In May, the delta variant was first diagnosed among a couple of patients in Texas. This more contagious version of the virus would quickly sweep through the United States and the world. The unvaccinated, which made up about 40% of the U.S. population by that time, were especially hard hit.
Infection rates among younger people were rising just as many locations had lifted lockdowns, made mask mandates optional and opened travel. Vaccine hesitancy was growing despite the predictions that the only real way to curb the rise in hospitalizations and deaths was for the population to get vaccinated.
Infection levels spiked again in September just as children around the country were heading back into classrooms, after nearly a year and half of virtual learning. A new variant, omicron, even more contagious that the delta variant, was identified in November. Both were circulating with speed through the global population.
“Each time we were presented with a question by different state and federal agencies, we tried to give the best answer possible for that question at that moment of time,” Marathe said. “Today, we are studying the waning immunity, vaccinations in children 5 to 12 years old, booster shots, and this new surge in cases to see how things might play out in the next three to four months. We’ve just focused on doing good work to support decision-makers.”
Crow, from the Virginia Department of Health, said the expeditions team has updated its models to account for variations in immunity levels, “which will allow us to understand the potential impact of new variants like omicron, which may evade immunity to one degree or another.”
The NSF’s Basu said the team’s ability to connect what’s happening locally to the global context is critical. “As a result, in five to 10 years, we will look back at the outcome of this research and say that ‘Yes, we have done something to address the global issue of pandemics.’”
Also baked into the goals of the expeditions grant is a focus on providing research opportunities for students. “Students are the key element here,” Basu said. “Often, they are more invested in the project because their livelihood depends on it, their thesis depends on it.”
Lijing Wang, a former Ph.D. student co-advised by Marathe, now a postdoc fellow at Boston Children’s Hospital and Harvard Medical School, focused on flu modeling and forecasting prior to the pandemic. “Once the pandemic started, I began to realize that it was necessary to check the state-of-the-art deep learning methods for epidemic forecasting with an emphasis on improving the way we explain the model outcomes to public health policy-makers so they could make better informed decisions,” Wang said.
Ann Li, a second-year UVA computer science student, is just getting started in research. “Working on this project has allowed me to explore the different aspects of research and appreciate its significance,” Li said. “With our project, I got to see firsthand how computer science research can be integrated into the real world, and its role in interdisciplinary systems such as public policy and epidemiology.”
The Biocomplexity Institute also recently earned a grant from UVA’s Prominence-to-Preeminence STEM Targeted Initiatives Fund for a project titled PREPARE – Pandemic Research in Emergence, Planning, and Response –seeking to reduce the global burden of infectious diseases through technology and engineering.
“UVA has provided incredible support to us as we worked on responding to the pandemic,” Marathe said.
Today, the expeditions team continues to advance the understanding of real-time epidemiology, with more than 70 peer-reviewed articles published, hundreds more presentations and the continued monitoring of a disease that continues its deadly march around the world.
“This disease is not going to go away in the foreseeable future, and the economic impact and long-term health impacts will be with us for a very long time,” Marathe said.
“This superb team of multi-disciplinary scientists, coming together to address an important societal problem, is providing the ability to capture information in close to real time about virus spread around the world, and has played a pivotal role in understanding where this virus is going and interventions that could have an impact on human life. We felt we had an opportunity to push the boundaries of computer science by looking at problems in epidemiology. And in doing that, also push epidemic science too. NSF empowered us to do science that really matters, and that has been our goal all along the way.”