How can synthetic data be generated, validated, and used to study the transition from sporadic infections to outbreaks when real-world data are sparse, biased, delayed, incomplete, or inaccessible?

Paper outlining synthetic dataset development, implementation, validation

Can changes in population-level and spatially clustered immunity be used to develop an early-warning signal for the transition from predictable seasonal disease dynamics to explosive outbreaks?

Report or model

An outbreak usually begins with sparse, delayed, and incomplete observations from a small number of locations. By the time conventional time-series indicators clearly show sustained growth, opportunities for early intervention may already be disappearing. Could changes in the structure and activity of the underlying transmission network provide earlier warning that an outbreak is shifting from localized transmission toward regional or wider expansion?

We propose studying outbreaks as dynamic, directed, multiscale graphs. Nodes could represent geographic areas, populations, healthcare systems, or host groups. Time-varying edges could represent mobility, contact, transportation, inferred transmission, or genetic relatedness. Signals observed on the nodes could include cases, hospitalizations, wastewater measurements, syndromic surveillance, and genomic indicators.

The working group would identify structural-temporal warning signals—such as changes in connectivity, directionality, community boundaries, or bridge populations—that precede expansion. A central question would be whether these signals transfer across pathogens, locations, spatial resolutions, and surveillance systems. Retrospective studies would simulate real-time emergence by restricting the information available at each point in an outbreak.

Graph-based approaches would be evaluated against time-series-only methods, static-network approaches, established epidemiological models, and ensemble forecasts. The group would explicitly study missingness, reporting delays, changes in surveillance, uncertainty, and negative transfer: situations in which a pattern learned from one outbreak produces misleading conclusions in another.

We seek interpretable principles for when network structure adds actionable information, when evidence from previous outbreaks can be reused, what additional observations would reduce uncertainty, and when a model should abstain from issuing an alert.

A multidisciplinary definition and taxonomy of network-based pandemic-expansion warning signals.

An open benchmark spanning multiple pathogens, locations, outbreak phases, and surveillance modalities.

An evaluation protocol that prevents future information from leaking into retrospective analyses.

Statistical and graph-learning methods for early warning, transferability assessment, uncertainty quantification, and detection of negative transfer.

An open-source diagnostic toolkit for outbreak analysts.

Academic publications and novel hypotheses about how network changes precede geographic expansion.

A policy and technical brief for public-health forecasting and surveillance teams.

A foundation for a larger interdisciplinary grant proposal.

The project could help public-health agencies determine when changes in mobility, transmission, or surveillance networks warrant heightened monitoring or intervention. It could also provide evidence-based criteria for deciding whether an early-warning model developed for one outbreak is safe to reuse in another setting.

Coinfection is widespread across disease ecology systems, yet empirical studies report highly variable and often inconsistent relationships among coinfecting parasites. Beyond genuine biological variation among systems, this inconsistency may also reflect substantial heterogeneity in how coinfection is observed and analyzed. Existing datasets vary widely in sampling design and resolution, ranging from presence–absence records to quantitative measures of infection intensity, while analytical approaches often rely on relatively standard statistical assumptions that may not reflect the underlying biological processes. As a result, potentially important coinfection relationships may be overlooked or difficult to interpret.

Mechanistic theory offers an opportunity to distinguish biological variation from methodological limitations by identifying plausible relationships that may otherwise be missed in empirical analyses. Our recent work provides a proof of concept, showing that theory can guide the detection of nonlinear coinfection relationships that become obscured under reduced data resolution or conventional analytical assumptions. Building from this foundation, we propose to systematically examine how coinfection data are structured and analyzed across disease ecology systems, determine how these differences affect what biological relationships can be detected, and develop a flexible, theory-informed framework for extracting greater biological insight from existing data.

Ultimately, we aim to translate these insights into a computational framework in which mechanistic knowledge prioritizes biologically plausible hypotheses while empirical data determine which relationships are supported, making data-driven and machine-learning analyses more efficient and interpretable.

A systematic synthesis of coinfection data structures and analytical approaches across disease ecology systems; a theory-informed framework for evaluating what biological relationships can be detected from different types of existing data; and a prototype computational approach for integrating mechanistic knowledge into data-driven analyses. These products will provide practical guidance for extracting greater biological insight from existing coinfection datasets and identify priorities for future empirical and theoretical studies.

By improving our ability to detect and interpret coinfection interactions across disease ecology systems, this framework could provide important insights for future disease control and management. Understanding when coinfecting pathogens facilitate or inhibit one another, and how these relationships change across ecological contexts, could inform whether and how targeting one pathogen may affect another. Translating these insights into specific management actions will require further collaboration with disease managers and other stakeholders.