Temporal dynamics of microbial communities in anaerobic digestion: Influence of temperature and feedstock composition...

This study presents a novel multi-disciplinary approach, integrating explainable machine-learning with detailed chemical and biological analysis using microbial fermentation food wastewater to identify critical parameters influencing AD microbiome diversity and reactor performance over time. Our research addresses knowledge gaps on the resource recovery potential from food fermentation wastewater which remains largely unexplored. Eighteen continuous bioreactors were operated under varying temperature and process configuration (single vs. two stages) using wastewater stream derived from real-world microbial protein manufacturing. Detailed chemical analysis fingerprinted the fermentation-wastewater throughout the AD process including sugars, sugar alcohols and volatile fatty acids (VFAs). Significant elevations in Methanomassiliicoccus were correlated to high butyric acid concentrations (p < 0.05) and decreased biogas production, further elucidating the role of this newly discovered methanogen in AD. Machine-learning models predicting reactor performance achieved high accuracy based on operational parameters and microbial taxonomy (14.03 root mean squared error and 0.86 R²). Operational parameters were found to have the most substantial influence on chemical oxygen demand removal. Oscillibacter and Clostridium sensu stricto were highlighted as key factors impacting specific biogas yield for the first time. By integrating detailed chemical and biological fingerprinting with explainable machine learning models this research presents a novel approach to advance whole-systems AD understanding, offering insights of potential new biomarkers (Oscillibacter and Clostridium sensu stricto) for industrial applications of sustainable waste-to-energy systems.