Predictive modeling of urinary 8-hydroxy-2´-deoxyguanosine (8-OHdG) in metal workers: A comparative study of machine learning algorithms in toxicology

Abstract

Urinary 8-hydroxy-2´-deoxyguanosine (8-OHdG), a biomarker for oxidative DNA damage, is commonly used to assess the repair of reactive oxygen species (ROS) induced DNA damage. This study developed predictive models to quantify 8-OHdG concentrations in urine samples based on demographic and exposure-related variables from metal workers (21.27 ng/ml) and controls (12.63 ng/ml) using ELISA and machine learning algorithms. Three models; Random Forest Regressor (RFR), Support Vector Machine Regressor (SVMR), and Gradient Boosting Regressor (GBR) were evaluated for their predictive performance using metrics like Mean Squared Error (MSE), Root Mean Squared Error (RMSE), R-squared (R2), Mean Absolute Error (MAE), and classification metrics including Accuracy, Precision, Recall, and F1 Score. The RFR emerged as the best regression model with an MSE of 1.35, RMSE of 1.16, R2 of 0.92, and precision of 0.89 where feature importance analysis indicated exposure and age as key predictors. The SVMR showed slightly lower performance (MSE = 1.54, R2 = 0.91, precision = 0.83). GBR had reduced regression performance (MSE = 1.66, RMSE = 1.29, R2 = 0.90) but achieved superior classification metrics all at 0.89. Overall, RFR provided the most accurate predictions, while GBR excelled in balancing classification performance. These findings indicated the efficiency of machine learning in quantifying oxidative stress biomarkers.