Congener-specific uptake and accumulation of bisphenols in edible plants: Binding to prediction of bioaccumulation by attention mechanism multi-layer perceptron machine learning model

Abstract

Plant accumulation of phenolic contaminants from agricultural soils can cause human health risks via the food chain. However, experimental and predictive information for plant uptake and accumulation of bisphenol congeners is lacking. In this study, the uptake, translocation, and accumulation of five bisphenols (BPs) in carrot and lettuce plants were investigated through hydroponic culture (duration of 168 h) and soil culture (duration of 42 days) systems. The results suggested a higher bioconcentration factor (BCF) of bisphenol AF (BPAF) in plants than that of the other four BPs. A positive correlation was found between the log BCF and the log Kow of BPs (R2carrot = 0.987, R2lettuce = 0.801, P < 0.05), while the log (translocation factor) exhibited a negative correlation with the log Kow (R2carrot = 0.957, R2lettuce = 0.960, P < 0.05). The results of molecular docking revealed that the lower binding energy of BPAF with glycosyltransferase, glutathione S-transferase, and cytochrome P450 (−4.34, −4.05, and −3.52 kcal/mol) would be responsible for its higher accumulation in plants. Based on the experimental data, an attention mechanism multi-layer perceptron (AM-MLP) model was developed to predict the BCF of eight untested BPs by machine learning, suggesting the relatively high BCF of bisphenol BP, bisphenol PH, and bisphenol TMC (BCFcarrot = 1.37, 1.50, 1.03; BCFlettuce = 1.02, 0.98, 0.67). The prediction of BCF for ever-increasing varieties of BPs by machine learning would reduce repetitive experimental tests and save resources, providing scientific guidance for the production and application of BPs from the perspective of priority pollutants.