A new perspective on real-time monitoring and function interpretation of nitrification based on machine learning and mass-to-charge fingerprint.

The stability of nitrification driven by nitrifying bacteria is susceptible to fluctuations in operating parameters. The traditional physicochemical parameters cannot directly reveal the metabolic activity of microorganisms; it remains essential to integrate direct characterization of microbial metabolic functional status for real-time monitoring of the variation during the nitrification process. Therefore, this study proposes a method that combines the proton-transfer-reaction time-of-flight mass spectrometry with machine learning. Based on 28,396 groups of mass spectrometry and water quality parameter data collected during the operation of the air-lift reactor, a real-time monitoring and analysis system centered on the mass-to-charge ratio (m/z) signals of microbial volatile organic compounds (mVOCs) during nitrification was constructed. XGBoost model demonstrates excellent analytical capabilities for key process parameters (test set R2: 0.96 for temperature, 0.95 for DO, 0.99 for pH) and specific characteristic m/z signals, particularly Mz42 and Mz87, show strong correlations with ΔNH₄⁺-N (R2 up to 0.83). This study further revealed the specific association between characteristic m/z signals and microbial metabolic states: an increase in Mz42 can indicate inhibition of nitrification (decrease in Nitrospira) and proliferation of Thiothrix, while Mz45 can effectively reflect Sphaerotilus proliferation. This study provides a new technical path for the paradigm shift of wastewater treatment monitoring from the traditional "water quality parameter monitoring" to "metabolic state monitoring".