Why PFAS?
PFAS are a large family of highly persistent chemical compounds associated with major environmental and public health concerns. Their spread, persistence, and complex exposure pathways require more integrated analysis methods.
Ongoing research project · ExposUM 2024
EAD
Environmental Anomaly Detection and Health Impacts: Addressing PFAS Risks
An interdisciplinary project combining environmental data analysis, knowledge representation,
anomaly detection, and visual analytics to better understand PFAS contamination and its potential impacts on ecosystems and public health.
PFAS
Ontology Engineering
Knowledge Graphs
Anomaly Detection
Visual Analytics
One Health
Project overview
PFAS are persistent pollutants detected in water, soil, wildlife, and human biological systems. EAD investigates how semantic technologies, machine learning, and visual exploration can be combined to detect anomalies, structure knowledge, and support explainable environmental analysis.
Core ambition
Build a comprehensive framework to integrate heterogeneous data, detect contamination patterns, and relate environmental exposure to potential health impacts.
Scientific perspective
The project follows a One Health vision, linking environmental quality, ecosystem dynamics, and human health through interoperable representations.
Main objectives
Knowledge representation
- Construct and enrich an ontology set dedicated to environmental pollutants with an emphasis on PFAS.
- Integrate structured and unstructured sources into a shared semantic model.
- Support interoperability, reasoning, and reuse through knowledge graphs.
Data-driven analysis
- Identify normal and abnormal behaviors through clustering and anomaly detection.
- Explore spatio-temporal methods and deep learning models for contamination analysis.
- Design interactive visual interfaces for transparent and explainable exploration.
Methodological pillars
Ontology engineering
Reuse and extend semantic resources to model pollutants, sources, exposure pathways, environmental effects, and health-related dimensions.
Hybrid anomaly detection
Combine clustering, deep learning, and semantic constraints to detect outliers, contamination zones, and emerging patterns in environmental datasets.
Visual analytics
Provide domain experts with interactive views, filters, and explainability features to inspect data, results, and uncertainty.
Current outcomes
Semantic resources
A first ontology and knowledge graph dedicated to PFAS and environmental exposure are available, following Linked Open Data principles.
Related resource: OntoPFAS
Exploration and explainability
A visualisation tool supports interaction with PFAS-related data, including geospatial inspection, filtering, and correlation analysis with explainability-oriented methods.
Clustering and interpolation
Initial work explores clustering approaches to mitigate sampling bias and improve trust in environmental analysis through confidence-aware spatial interpolation.
Towards enriched data hubs
The project aims to connect raw environmental measurements, structured semantic resources, and broader data platforms for more reusable PFAS knowledge infrastructures.
Project timeline
Phase 1 — Foundations
Literature review, documentation, team meetings, and project coordination.
Phase 2 — Ontology construction
Building and evaluating an ontology for environmental pollutants, with a focus on PFAS.
Phase 3 — Information extraction
Developing models to populate the ontology from structured and textual data sources.
Phase 4 — Anomaly detection and analytics
Designing data analysis approaches, contamination clustering strategies, and predictive models.
Phase 5 — Visual exploration and reporting
Building expert-oriented interfaces, final evaluation, and dissemination.
Consortium
Partners
- MISTEA (INRAE / Institut Agro)
- LIRMM (CNRS / Université de Montpellier / Paul-Valéry Montpellier)
- EPOC (CNRS)
- PRODIG (CNRS)
Scientific leadership
The project is led by Pascal Neveu and Lylia Abrouk, in collaboration with specialists in machine learning, chemistry, geography, environmental health, and data journalism.