Prediction of p-phenylenediamine antioxidant concentrations in human urine using machine learning models

p-phenylenediamine antioxidants (PPDs) are extensively used in rubber manufacturing for their potent antioxidative properties, but PPDs and 2-anilino-5-[(4-methylpentan-2yl)amino]cyclohexa-2,5-diene-1,4-dione (6PPDQ) pose potential environmental and health risks. Existing biomonitoring methods for assessing human exposure to PPDs are labor-intensive, costly, and provide limited data. Thus, there is a critical need to develop predictive models for evaluating PPDs and 6PPDQ exposure levels to facilitate health risk assessments. In this study, machine learning (ML) models were developed to predict the concentration of three PPDs and 6PPDQ in human urine samples. A total of 759 participants from three cities in Zhejiang Province, China, provided urine samples, which were analyzed for PPDs and 6PPDQ concentrations using liquid chromatography-tandem mass spectrometry. Eight ML models were employed to predict PPDs and 6PPDQ concentrations based on demographic and environmental exposure factors such as age, gender, body mass index (BMI), and occupation. N-(1,3-dimethylbutyl)-N'-phenyl-p-phenylenediamine (6PPD) was the most frequently detected PPD (mean 3.03 ng/mL, range < LOD-18.65 ng/mL), followed by 6PPDQ (mean 2.76 ng/mL, range < LOD-20.85 ng/mL) and N-phenyl-N'-cyclohexyl-p-phenylenediamine (mean 2.04 ng/mL, range < LOD-10.22 ng/mL). Random forest model demonstrated the highest accuracy in predicting PPDs and 6PPDQ concentrations in human urine among the ML models evaluated. Through the application of these models, age, BMI, and occupation emerged as significant predictors of urinary PPDs and 6PPDQ concentrations. This research significantly contributes by using ML models to enhance exposure assessment accuracy and efficiency, providing a novel framework for future studies on environmental health risks related to PPDs and 6PPDQ.