Coupling growth kinetics modeling with machine learning reveals microbial immigration impacts and identifies key environmental parameters in a biological wastewater treatment process

Abstract

Abstract Background Ubiquitous in natural and engineered ecosystems, microbial immigration is one of the mechanisms shaping community assemblage. However, quantifying immigration impact remains challenging especially at individual population level. The activities of immigrants in the receiving community are often inadequately considered, leading to potential bias in identifying the relationship between community composition and environmental parameters. Results This study quantified microbial immigration from an upstream full-scale anaerobic reactor to downstream activated sludge reactors. A mass balance was applied to 16S rRNA gene amplicon sequencing data to calculate the net growth rates of individual populations in the activated sludge reactors. Among the 1178 observed operational taxonomic units (OTUs), 582 had a positive growth rate, including all the populations with abundance > 0.1%. These active populations collectively accounted for 99% of the total sequences in activated sludge. The remaining 596 OTUs with a growth rate ≤ 0 were classified as inactive populations. All the abundant populations in the upstream anaerobic reactor were inactive in the activated sludge process, indicating a negligible immigration impact. We used a supervised learning regressor to predict environmental parameters based on community composition and compared the prediction accuracy based on either the entire community or the active populations. Temperature was the most predictable parameter, and the prediction accuracy was improved when only active populations were used to train the regressor. Conclusions Calculating growth rate of individual microbial populations in the downstream system provides an effective approach to determine microbial activity and quantify immigration impact. For the studied biological process, a marginal immigration impact was observed, likely due to the significant differences in the growth environments between the upstream and downstream processes. Excluding inactive populations as a result of immigration further enhanced the prediction of key environmental parameters affecting process performance.

Keywords

• Immigration impact
• Active population
• Machine learning