Due to the complexity of microbiome composition and interactions, it remains challenging to uncover the bioreactor "black-box" and to predict bioreactor dynamics (e.g., anaerobic digestion) through linking the microbial community, operational factors and system performances. Previous studies have focused on abundant/functional microorganisms and community diversity. The importance of microbial interactions and potential low-abundance members may be underestimated. Microbial interactions can be modeled through network analysis using the more and more accessible 16S gene sequencing data. However, pooling all samples (systems/stages) for one network model may not be the best approach. One major gap is how to build/interpret/translate network model result to linked with engineering system prediction.
Aim. This study aims at developing network models of AD microbiome and linking the model output with operational/performance parameters.
Methods.Microbial communities from 12 lab-scale AD reactors at different stages were collected with physical-chemical variables. Five groups of communities were clustered to build separate co-occurrence networks.
Findings. The five network models showed different topological properties. The hydrolysis efficiency correlated with Clustering Coefficient positively and Normalized Betweenness negatively. The Average Path Length correlated negatively withinfluent particulate COD and differential hydrolysis-methanogenesis efficiency. Individual OTUs' topological characteristics showed that high-ratesystem had more connectors,and low-abundance OTUs could perform central hub roles and communication roles, maintaining the stability and functionality. Our study revealed that stronger interaction is linked to higher rates/activities. The Average Path Length seems to have a global meaning in different microbiomes (activated sludge, soil), correlated negatively with substrate availability and utilization efficiency.
Utilisation. This study provides a framework to build network models of microbiome data and extract key indicators from model outputs to indicate system performance. The model could be customized and expanded for specific systems to generate links with engineering operations/performance, which could be used for system prediction and optimization.