Metabolomics-based estimation of activated sludge microbial composition and prediction of filamentous bulking

Abstract

Understanding the structure and activity of activated sludge (AS) microbiome is key to ensuring optimal operation of wastewater treatment processes. While high-throughput metagenomics offers a comprehensive view of AS microbiome, its cost and time demands warrant alternative approaches. This study employed machine learning methods to integrate metabolomic and metagenomic data, enabling predictions of selected microbial abundances from metabolite profiling. Model training relied on rich microbial and metabolite abundance data collected in an intensively sampled AS system, including a period of filamentous bulking, as well as a few other AS systems. Multiple linear regression out-competed other three algorithms in achieving relatively high prediction accuracy (R2 = 0.70±0.02) for the abundances of 10 selected, either keystone or core metagenome-assembled genomes (MAGs). The model predicted the abundances of filamentous Microtrichaceae and Thiotrichaceae during bulking with an error range of 14–17.8 %. This predictive power extends beyond the specific system studied, showcasing potentials for broader applications across other AS systems. Aspartate, glycine, and folate were the most influential metabolite features contributing to model performance, which were also effective indicators for filamentous bulking, with up to one week of early warning potential. This study pioneers the application of metabolomics for fast, relatively accurate and cost-effective prediction of AS community composition, enabling proactive management of AS systems towards improved efficiency and stability.