Answers are grounded in this paper and cite their sources.
Plant metabolite profiles offer an indirect approach to characterizing soil exposure to polystyrene particles. This study evaluated leaf-based, stem-based, and joint leaf–stem models for classifying four soybean exposure treatments: an untreated control and polystyrene treatments of 5 mm, 500 μm, and 25 μm at 0.1% soil loading. Four groups of 25 pots, each containing one plant, were established. Sample preparation losses left 64 paired leaf–stem records. These were divided into training (n = 51) and test (n = 13) sets. Outlier screening within the training set retained 47 paired records, which formed the joint training dataset. H2O AutoML was used to develop classification models with approximately 80% training data and 20% held-out test data, with five-fold cross-validation within the training datasets. The leaf-based, stem-based, and joint GBM models achieved mean cross-validation accuracies of 80.9% ± 4.6%, 80.4% ± 12.1%, and 93.6% ± 5.9%, respectively. Corresponding test accuracies were all 92.3% (12/13). The joint model achieved the highest observed accuracy, although the small and differing test datasets limited direct comparison. Variable importance identified tissue-specific predictors rather than statistical significance or causal mechanisms. These findings provide a proof of concept for classifying controlled polystyrene exposure treatments at the tested loading, but performance at environmentally relevant concentrations and under field conditions remains to be established.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: