Inferring nitrifying taxa abundance in activated sludge from treated wastewater: a data-driven approach

Abstract

Effective control of wastewater treatment plant (WWTP) performance requires monitoring of key parameters that reflect biological process stability. Increasingly, next-generation sequencing (NGS) is used to characterize microbial communities involved in nitrogen removal; however, most studies focus on activated sludge (AS), often overlooking treated wastewater (TW), which also contains living microorganisms and may provide valuable diagnostic information. In this study, the microbiomes of raw sewage (RS), AS, and TW from a full-scale biological WWTP were compared across multiple taxonomic levels. The potential of TW as a predictive indicator of AS microbiome composition, with particular emphasis on nitrifying bacteria responsible for nitrogen removal, was evaluated. In addition, relationships between microbial communities at different treatment stages and environmental factors were analysed. The results indicate that data-driven approach can successfully model the abundance of specific AS nitrifying taxa based on TW data, achieving an R 2 score of 0.9826 on the training dataset and maintaining a high predictive accuracy with an R 2 above 0.9996 on the testing dataset. This study demonstrates the feasibility of utilizing TW microbiome data as a predictive proxy to infer the abundance dynamics of key nitrifying functional groups in AS via explainable machine learning. The proposed proof-of-concept introduces an explainable machine-learning-based approach to monitoring nitrogen-removing microbial functional groups, with potential to support WWTP operation and optimization.