Abstract
Plant uptake and accumulation of per- and polyfluoroalkyl substances (PFAS), represented by the root concentration factor (RCF), shoot concentration factor (SCF), and translocation factor (TF), were predicted using machine learning (ML) models from experimental data with 19 PFAS compounds and nine plant species. Unsupervised principal component analysis (PCA) was first used to classify the input data, and then supervised ML models, including multiple linear regression model (MLR), artificial neural network (ANN), random forest (RF), and support vector machine (SVM) algorithms, were applied for predicting the chosen output parameters. RF displayed the highest prediction accuracy among the tested models. Feature importance analysis performed using RF showed that the molecular weight, exposure time, and plant species are the most important parameters for predicting RCF, SCF, and TF in hydroponic systems. RF was further applied to estimate RCF, SCF, and TF of the two most prevalent PFAS compounds, perfluorooctanoic acid (PFOA) and perfluorooctanesulfonic acid (PFOS), and their common alternatives and the results revealed that their common replacing compounds have either comparable or higher accumulation in plant roots and shoots. Our results demonstrated that the ML approach could generate critical insight into PFAS plant uptake and accumulation and shed light on the potential food safety concerns from PFAS and their replacements.