Predicting wastewater effects on antibiotic fluorescence from laboratory DOM quenching behavior by fingerprinting and machine learning

Abstract

Antibiotic contamination in environmental waters poses both ecological and health hazards. However, detecting antibiotics using fluorescence techniques is challenging due to pH-dependent spectral variability and interference from dissolved organic matter (DOM). In this study, the excitation–emission matrix (EEM) fluorescence spectroscopy combined with parallel factor analysis (PARAFAC) was used to characterize the intrinsic fluorescence properties of 27 antibiotics. Based on environmental prevalence and structural class, antibiotics were systematically selected at three environmentally relevant pH values (5, 7, and 9) that reflect the typical pH range found in natural and wastewater systems. Results revealed strong stability in fluoroquinolones, pH-dependent enhancement in tetracyclines, and negligible emission in several other classes. Based on these results, Random Forest classifiers trained on 19 spectral features achieved 85.2% accuracy for pH response prediction and 92.6% for detection feasibility. Based on these findings, we developed a Detection Risk Index (DRI) that categorizes 44% of antibiotics as low risk, 33% as medium risk, and 22% as high risk for EEM-based detection. Five antibiotics were selected for DOM interaction and wastewater validation studies based on their DRI and environmental relevance. Experimentation with DOM interaction patterns showed that there is a wide range of quenching behavior among antibiotics, ranging from dynamic quenching to fluorescence enhancement. Notably, the directional consistency between laboratory DOM quenching at neutral pH and fluorescence matrix effects in wastewater indicates that controlled experiments can predict environmental interference. The results provide valuable information for monitoring antibiotics using fluorescence techniques in environmental waters.