“Forever chemicals,” also known as per- and polyfluoroalkyl substances (PFAS), are a class of potentially toxic, manmade compounds that contaminate drinking water and environmental systems across the United States. Although PFAS compounds are used widely and pose varying levels of environmental exposure, toxicity data for PFAS are limited, and many human hazards of PFAS exposure are unknown.

We developed AI4PFAS, a suite of machine learning models for assessing the nonspecific toxicities of PFAS compounds, to address the lack of PFAS toxicity data. The AI4PFAS workflow is uncertainty-informed via a selective prediction paradigm. The tools built for this workflow will continue to guide high-throughput screening of environmental contaminants.