Public health surveillance systems continually face significant hurdles when attempting to integrate fragmented health records, environmental observations, and societal activity streams into an actionable epidemiological overview. To tackle these structural data barriers, researchers have designed and demonstrated a three-layered semantic architecture that organizes multi-domain pandemic intelligence into interconnected knowledge graphs. Reporting in the Journal of Biomedical Semantics, the investigative team introduced an ontology-driven platform capable of bridging disparate structured tables and unstructured documents. By transforming isolated reports into a coordinated semantic web, the framework establishes computational pathways between clinical outcomes and the complex real-world circumstances surrounding pathogen transmission, allowing analytical systems to detect meaningful epidemiological correlations that typically remain buried within siloed archives.
The necessity for comprehensive contextual integration has grown acutely visible during international disease emergencies, where conventional monitoring systems often fail to capture broader societal determinants. In 2020 alone, the World Health Organization coordinated responses to 116 public health crises spanning 194 locations worldwide, illustrating the relentless pressure placed upon administrative decision-makers. Traditional surveillance programs have historically gathered routine clinical notifications, yet they frequently lack the capacity to incorporate vital contextual information regarding human movement, mass gatherings, and community encounters. Although established resources like the Global Outbreak Alert and Response Network, MedISys, and the Global Public Health Intelligence Network have made substantial contributions by tracking open media and incident feeds, stakeholders still confront disparate analytical tools and fragmented datasets that complicate timely intervention planning.
To overcome these challenges, the investigators grounded their approach in semantic web technologies, constructing a unified knowledge representation system anchored by formal ontologies. Ontologies supply machine-readable vocabularies of defined classes, attributes, and formal logical axioms, establishing a coherent structure across heterogeneous information streams. Rather than relying on a solitary, rigid ontology that would prove difficult to maintain, the research group implemented a modular three-layered hierarchy. This multi-tiered paradigm separates general abstractions from granular operational parameters, fostering reusability, modular expansion, and long-term consistency. The design ensures that baseline public health concepts remain standardized while allowing specialized domain models to articulate intricate real-world phenomena without introducing structural disorder across the broader ecosystem.
At the apex of this architectural model sits the Open Reference Ontology, which serves as a foundational abstraction layer defining fundamental concepts common across all public health operational scenarios. Developed specifically to capture the overarching mechanisms of biological threats, the Open Reference Ontology formally models core entities including hazards, threats, risk factors, vulnerabilities, diseases, preventive measures, mitigations, and systemic impacts. It establishes governing semantic relationships such as causes, effects, mitigates, prevents, and source of. By formulating these primary predicates, the reference tier functions as a universal backbone, enabling subordinate domain-specific ontologies to inherit foundational logic and compose contextual information through shared derivation rules and standardized knowledge constraints.
Beneath the reference framework operates the intermediate tier, comprising 16 domain-specific ontologies meticulously assembled through direct collaboration with subject-matter experts. These ontologies articulate specialized operational areas including pathogens, infectious diseases, symptom expressions, chemical agents, geographic entities, healthcare operations, passenger travel, and mass gathering events. To preserve interoperability and adhere to established Open Biological and Biomedical Ontology Foundry principles, the researchers extensively reused existing biomedical resources accessed via the Ontology Lookup Service and Ontofox. Concepts were incorporated from standard repositories, including the National Cancer Institute Thesaurus, Chemical Entities of Biological Interest, Human Disease Ontology, National Center for Biotechnology Information Taxonomy, Symptoms Ontology, Human Phenotype Ontology, and GeoNames for geographic representation.
The lowest tier consists of application-specific ontologies, tailored to align specific empirical datasets directly with the broader domain schema. For their primary operational implementation, the researchers developed an application ontology mapped to the World Health Organization Europe COVID-19 line list dataset stored within the secure, non-public WHO xMART platform. This specialized application ontology meticulously captures clinical details, including patient hospital admission and discharge schedules, clinical signs, symptoms, comorbidities, lab test diagnostics, exposure locations, and individual travel trajectories. By applying formal Web Ontology Language restrictions—including specific value constraints, cardinality parameters, and class inheritances—the application tier binds individual case files directly to broader epidemiological definitions, creating an expressive semantic bridge for deep analysis.
Constructing this extensive digital architecture required a coordinated technical pipeline utilizing specialized authoring and management platforms. The investigators built their ontologies within the Protégé modeling environment using the Web Ontology Language accompanied by Resource Description Framework Schema annotations for documentation. Unstructured narrative feeds, primarily gathered from Epidemic Intelligence from Open Sources documents, were processed using the automated text-mining tool SCAIVIEW, which recognized target concepts and converted textual entities into structured Resource Description Framework triples governed by Simple Knowledge Organization System rules. These generated triples, alongside structured tables converted from spreadsheet and database formats, were deposited directly into an Apache Jena Fuseki server and persistently archived using Jena’s native TDB triple store database.
A critical analytical advantage demonstrated by the framework is its capacity to perform automated inference using an integrated reasoning engine. Operating with the Resource Description Framework Schema Rule Reasoner, the system evaluates subclass and subproperty hierarchies alongside datatype range validations to derive implicit assertions that are not explicitly stated in the source files. In an experimental demonstration, the researchers examined the knowledge profile of an individual recorded simply as a patient within the raw clinical records. Under baseline querying without reasoning, the platform only retrieved this direct patient assertion; once the inference reasoner was engaged, the interconnected ontologies deduced that this individual was also a person who had participated in a mass gathering, had been exposed to an active hazard, and was directly relevant to public health containment protocols.
To evaluate the practical capabilities of the framework in a realistic historical scenario, the investigators deployed their system across a complex multi-domain case study centering on early European coronavirus transmission in February 2020. The evaluation examined the UEFA Champions League football match between Atalanta and Valencia held in Bergamo, Italy, an event highlighted by epidemiological analyses as a significant factor in accelerating subsequent regional outbreaks during March and April of that year. The research team assembled disparate data sources, including the WHO xMART patient line list, EIOS open-source reports, FlightAware international flight tracking records, stadium schedules, and GeoNames spatial coordinates, demonstrating how previously separate datasets could be jointly navigated through unified semantic querying.
Through targeted SPARQL protocol queries executed across the Jena Fuseki triple store, the framework successfully linked heterogeneous data streams into a cohesive property graph visualized within Neo4j. The automated anomaly detection algorithms leveraged these integrated semantic graphs to identify structural connections between news reports detailing emerging viral symptoms, the scheduled sports fixture in Bergamo, and specific airline flight paths departing and arriving across associated municipal transit hubs. By formalizing entities such as the Champions League under the Mass Gathering ontology and linking them to geographic and transport vectors, the architecture proved that multi-source surveillance data can be reconciled to illuminate contextual transmission patterns that isolated databases inevitably obscure.
Despite these demonstrated strengths, the authors documented several technical constraints and logistical trade-offs inherent in large-scale semantic web deployment. Manually mapping local application ontologies to broad domain models remains an intensive, time-consuming process requiring substantial manual oversight from domain specialists and data curators. Furthermore, semantic web tools continue to evolve, requiring specialized technical training for healthcare practitioners and data providers before institutional adoption can succeed. The researchers also noted that under project time limitations, contextual classifications were occasionally hardcoded; for example, mass gatherings were categorized directly under hazards to accommodate the specific outbreak use case, whereas in production environments such classifications require conditional threshold evaluations.
Looking forward, the authors underscored that ongoing management of evolving ontology versions and expanding domain vocabularies will be critical for sustained public health operations. While the prototype demonstrated successful tracking around the Bergamo sporting event, expanding the system to monitor alternative biological hazards—such as West Nile virus transmission or wastewater-based disease detection—will require introducing dedicated sub-properties and refining the Open Reference Ontology with distinct biological and chemical hazard classifications. Nevertheless, by demonstrating how semantic enrichment, automated rule-based reasoning, and linked data principles can synthesize structured line lists with open news and transit matrices, the study establishes an extensible blueprint for next-generation epidemic intelligence systems.
Subject of Research: Semantic web ontology frameworks for multi-domain public health surveillance and pandemic intelligence
Article Title: Three-layered semantic framework for public health intelligence
Article References: Guru Rao, S., Rokkam, P., Zhang, B., Sargsyan, A., Kaladharan, A., Sethumadhavan, P., Jacobs, M., Hofmann-Apitius, M., & Tom Kodamullil, A. (2025). Three-layered semantic framework for public health intelligence. Journal of Biomedical Semantics, 16(1), Article 17. https://doi.org/10.1186/s13326-025-00338-1
Image Credits: AI Generated
DOI: 10.1186/s13326-025-00338-1
Keywords: public health intelligence, disease surveillance, semantic framework, ontologies, knowledge graphs, linked data, COVID-19, RDF triples, Apache Jena Fuseki, Protégé, epidemic intelligence, EIOS, mass gatherings, WHO xMART, FlightAware, SPARQL
Cite Scienmag News
Kristina Jarvis. (November 19, 2026). Three-Layered Semantic Framework Links Fragmented Surveillance Data for Pandemic Intelligence. Scienmag. https://scienmag.com/three-layered-semantic-framework-links-fragmented-surveillance-data-for-pandemic-intelligence/
Kristina Jarvis. "Three-Layered Semantic Framework Links Fragmented Surveillance Data for Pandemic Intelligence." Scienmag, 19 November 2026, https://scienmag.com/three-layered-semantic-framework-links-fragmented-surveillance-data-for-pandemic-intelligence/. Accessed 19 November 2026.
Kristina Jarvis. "Three-Layered Semantic Framework Links Fragmented Surveillance Data for Pandemic Intelligence." Scienmag. November 19, 2026. https://scienmag.com/three-layered-semantic-framework-links-fragmented-surveillance-data-for-pandemic-intelligence/








