Microbial Community Biomarkers Can Forecast Methane Production in Full-scale Anaerobic Digesters

Methane production from anaerobic digestion of wastewater sludge is complex, with disruptions in microbial stages potentially leading to system failure. Early detection of disturbances is crucial. While machine learning has been applied to methane prediction, no studies have used metagenomic or transcriptomic data as predictor variables. We employed random forest analysis to forecast methane production in three full-scale sludge digesters at a Singapore wastewater treatment plant. Over 25 weeks, we measured 42 physicochemical variables and performed shotgun metagenome and total RNA transcriptome sequencing. Recursive feature elimination identified reactor-specific predictors, largely representing rare and digester-specific microbial taxa. When data from all digesters were combined, key predictors included both physicochemical parameters, such as chemical oxygen demand, and microbial taxa. Learning curve simulations suggested that increasing sample size from 75 to 150–200 would improve prediction accuracy. Notably, many microbial predictors were unidentified operational taxonomic units, highlighting the role of yet-unknown microorganisms in anaerobic digestion. The resulting model is practical for onsite digester monitoring, requiring only the identified predictor variables to be measured