Mitigating ‘forever chemicals’ faster with AI and novel modeling techniques
Breakthrough framework, risk prediction map work in tandem
Widespread contamination from per- and polyfluoroalkyl substances (PFAS), often called “forever chemicals,” is a major public health concern that scientists at Los Alamos National Laboratory are addressing with two new tools.
The team created a national-scale risk prediction map that can aid researchers and communities prioritize monitoring and remediation efforts. They also developed a breakthrough framework that combines machine learning, molecular dynamics simulations and experimental analysis to understand how fast PFAS molecules can move through water, soils and sediments.
Read the paper
Why this matters: By integrating these approaches, scientists can better forecast PFAS behavior in groundwater and the environment, as well as accelerate the development of remediation strategies tailored to PFAS’ unique persistence.
- PFAS resist degradation, persist in soil and water, and accumulate in human and ecological systems.
- The compounds have links to some cancers and other diseases, immune system dysfunction and reproductive harm.
How they did it:
- For the risk prediction map, researchers built machine learning models that can integrate national geospatial datasets with environmental and industrial information such as proximity to PFAS-related facilities (manufacturing, military sites, landfills), hydrological connectivity, and soil and geological features that influence contaminant movement.
- For the fate and transport model, researchers trained a regression-based approach by iteratively targeting the most uncertain predictions for additional training. This data-efficient and adaptive process significantly improved model accuracy, reducing prediction error by 88% in estimating the physicochemical properties of PFAS.
Funding: The Laboratory Directed Research and Development program and the U.S. Department of Energy’s Office of Environmental Management
LA-UR-25-31133
