Building Predictive Understanding of the Activated Sludge Microbiome by Bridging Microbial Growth Kinetics and Microbial Population Dynamics

Modeling microbiomes can provide predictive insights into microbial ecology, but current modeling approaches suffer from inherent limitations. In this study, a novel modeling approach was developed based on the intrinsic connection between the growth kinetics of guilds and the dynamics of microbial populations. To implement this approach, 466 samples from four full-scale activated sludge systems were retrieved. The raw samples were processed using a data transformation method that tripled the data set size and enabled quantification of population dynamics. Of the 42 family level core populations, 36 showed overall dynamics statistically close to zero (within ± 0.05 d^-1). Bayesian networks were built to classify the core populations into heterotrophic and autotrophic guilds. Topological data analysis was applied to identify keystone populations and time-dependent microbial interactions. The data-driven inferences were validated directly using the Microbial Database for Activated Sludge (MiDAS) and indirectly by predicting community structure using artificial neural networks. The Bray-Curtis similarity between predicted and observed communities was higher with microbial kinetic parameters than without these parameters (0.70 vs 0.66, <i>t</i> test, <i>p</i> < 0.05). Owing to the flexibility of the modeling framework, this proposed hybrid approach might potentially be adapted to time-dependent data from natural systems for predictive understanding of the involved microbiomes.