Predicting Bioconcentration Factors and Concentrations of Per- and Polyfluoroalkyl Substances in Plants with Machine Learning
Year: 2026
Type: article
Abstract: Bioconcentration factors (BCF) and concentrations of per- and polyfluoroalkyl substances (PFAS) in crops were predicted by four machine learning models using a newly compiled data set composed of 11 crops, 11 plant tissue compartments, and 21 PFAS, as well as diverse soil properties and cultivation conditions. The eXtreme Gradient Boosting (XGBoost) achieved the best performance, with a coefficient of determination (R2) of 0.951 for Log BCF and 0.943 for Log C. Feature importance analysis showed that soil pH and PFAS carbon chain length are key predictors of BCF. Soil PFAS concentration is the primary input feature for PFAS concentrations in plants and shorter growth-cycle plants accumulated higher PFAS concentrations. Root and leaf tissues were distinguished as the primary compartments of PFAS accumulation, exhibiting the highest BCF and concentration levels. Interestingly, although PFAS accumulation in plants generally shows a negative correlation with carbon chain length, root accumulation has a relatively weaker correlation with high levels of accumulation for both short-chain and long-chain PFAS. In contrast, above-ground tissues (particularly leaves) preferentially accumulate short-chain PFAS. These findings could provide valuable insights into the development of management and mitigation strategies to minimize PFAS accumulation in food crops and subsequently human exposure through the food chain.
Authors: Rui Li, Jiyoon Yi, Hui Li, Wei Zhang
Institution: Michigan State University
Language: English
Subfield: Environmental Chemistry
Field: Environmental Science
Domain: Physical Sciences
SDGs: Zero hunger, Life on land
Open Access status: hybrid
Estimated APC paid: $4,500
OpenAlex ID: W7207610906