AI Researcher Wins NSF CAREER Award to Sharpen Epidemic Forecasting

Forecasting the next outbreak may soon depend less on a single stream of case reports and more on artificial intelligence systems capable of weaving together many imperfect signals at once. At the University of Virginia’s School of Engineering and Applied Science, assistant professor of computer science Chen Chen is building exactly that kind of system, an approach to epidemic intelligence that fuses reported infections, wastewater surveillance and human mobility data into forecasts designed to tell public health officials not only what is happening now, but what is likely to happen next. The project, titled Building Next-Generation Epidemic Intelligence: Forecasting, Intervention, and Surveillance, has earned Chen a five-year, $600,000 Faculty Early Career Development Program award from the U.S. National Science Foundation, the agency’s most prestigious honor for early-career researchers who show promise as both scholars and educators.

Chen’s route to computational epidemiology runs through a seemingly unrelated field: the study of connectivity in complex networks. Her doctoral research produced a unified framework for measuring and understanding how connectivity behaves across very different kinds of systems, from infrastructure networks such as power grids to the networks through which epidemics move. The unifying insight is that in any such system, the behavior of the whole depends on how its individual nodes are linked. In a power grid, engineers ask whether the remaining parts of the network can keep electricity flowing when a single node fails. In an epidemic, the nodes are people, and the connections between them, to co-workers, family members and strangers encountered in transit, become the pathways along which a pathogen travels through a geographic area.

That framing makes the difference between a contained outbreak and a pandemic a question of network dynamics. When transmission expands exponentially and crosses international boundaries, the same mathematical intuition that describes cascading failures in infrastructure applies to chains of infection. Chen describes her epidemic work as a natural extension of her dissertation idea, noting that people are connected to their co-workers and their families, and that the research aims to develop strategies for monitoring a population-level outbreak at different scales, with the ultimate goal of containing transmission and reducing the number of infections.

The technical core of the project addresses a well-known weakness in existing epidemic models. Forecasting and tracking systems can miss important pieces of the picture, capturing reported infections without fully incorporating how people move between communities or what wastewater monitoring suggests about infections that have never been reported. Mobility matters enormously, Chen explains, because someone traveling frequently from state to state establishes connections with many different people, and those connections can pose infection risk far from home. Wastewater surveillance adds another complementary signal, detecting viral traces shed by infected individuals regardless of whether they have sought testing, which became especially valuable during the COVID-19 pandemic when reported case counts were distorted by uneven testing and widespread home testing that never entered official statistics.

Bringing these heterogeneous data streams together in real time is a problem that only recently became tractable. Chen, a former Google software engineer, says the emergence of AI-powered analytical tools is what now allows the kind of data processing these projects require. Her industry experience shapes how she approaches the work: it is not enough to invent new algorithms, she argues, because the harder question is whether those algorithms can scale to large, complex datasets and ultimately prove useful in real-world settings. Her research interests in data mining, machine learning, computational epidemiology, and trustworthy and efficient AI converge on that goal, aiming to serve both scientific discovery and practical public health decision-making.

Trustworthiness is not an abstract virtue in this context. In high-stakes applications such as public health, Chen notes, AI models must account for uncertainty and potential errors and must produce results that researchers and decision-makers can understand and rely on. A forecast that cannot explain its own confidence, or that silently fails when conditions shift, is worse than no forecast at all when officials are deciding where to deploy limited resources. This requirement for interpretable, uncertainty-aware predictions is a recurring theme in the broader movement toward trustworthy AI, and Chen’s project embeds it directly into the design of epidemic forecasting tools rather than treating it as an afterthought.

Adaptability presents another technical challenge, because the pathogen itself does not stand still. The emergence of new variants during COVID-19 demonstrated how quickly the characteristics governing a disease and its transmission can change, potentially invalidating models trained on earlier conditions. Guanghui Min, a Ph.D. student working with Chen on the project, describes one exciting next step as combining different signals, such as case counts, mobility patterns and wastewater data, to obtain a more complete picture of how an outbreak is evolving. Min adds that the team believes making these models adapt when conditions change, so that forecasts can help identify where additional monitoring or intervention is needed, may be the most useful capability of all. In machine learning terms, this is a problem of distribution shift: the statistical relationships a model learned during one phase of an epidemic may no longer hold in the next.

The CAREER project is organized around several components: improving epidemic forecasting, developing computational tools to help evaluate possible interventions, designing better disease-surveillance strategies, and educating the public. Chen plans to concentrate heavily on forecasting during the first three years, creating a foundation for the intervention and surveillance tools that follow. The practical questions those tools are meant to answer are deceptively simple: What is happening now? What is likely to happen next? And if officials intervene, how might the trajectory change? Answering them requires models that can simulate counterfactuals, estimating how a vaccination campaign, school closure or targeted testing effort might bend the curve in one community while leaving another largely untouched.

Scale is the central obstacle between detailed local knowledge and useful national guidance. Inside a hospital, researchers might map interactions between infected individuals in fine detail; across a city, state or country, that becomes far more difficult. Chen’s framework is designed to work gracefully with different levels of available information. Publicly accessible disease reports, wastewater measurements and mobility data can provide a broad view of spread, and when more detailed records such as localized case counts, testing data and other health surveillance information become available for research, the framework can incorporate them to yield a more granular understanding of transmission and sharper predictions. The same logic points toward smarter surveillance: if a region lacks sufficient information, a model might show officials where an additional wastewater monitoring point would most improve forecasts, turning the monitoring network itself into an optimization problem.

Beyond the models and decision-support tools for officials, Chen envisions educational tools for the public, with a long-term goal of an accessible website that helps community members understand disease conditions and risk in the places around them, answering questions such as what the viral level is in a nearby area and what the risk level might be for someone visiting. Sandhya Dwarkadas, Walter N. Munster Professor and chair of the Department of Computer Science, observes that these are interesting times, with access to unprecedented amounts of data, analytical tools and real-world insight from COVID-19 about how to apply them, and that she appreciates how Chen is maximizing that advantage to help make a difference for communities worldwide by informing public health decision makers. The framework itself is not tied to any single pathogen: by changing how a model represents transmission and contact patterns, Chen says, the same approach could potentially be adapted to different infectious diseases. For Chen, the ultimate measure of success is whether better computational models can turn scattered information into useful guidance, helping officials respond to epidemics, prevent pandemics and give individuals a clearer picture of the risks around them.

Subject of Research: Artificial intelligence for infectious disease epidemic forecasting and surveillance

Article Title: Chen’s NSF CAREER Award supports UVA Engineering artificial intelligence research to better forecast infectious disease epidemics

Article References: Chen’s NSF CAREER Award supports UVA Engineering artificial intelligence research to better forecast infectious disease epidemics. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: epidemic forecasting, artificial intelligence, NSF CAREER Award, wastewater surveillance, public health, machine learning, network science, disease surveillance, human mobility data, computational epidemiology, trustworthy AI, University of Virginia