Decoding the emergent properties of microbial community functions through sub-community observations and interpretable machine learning

The functions of microbial communities, including substrate conversion and pathogen suppression, arise not as a simple sum of individual species’ capabilities but through complex interspecies interactions. Understanding how such functions arise from individual species and their interactions remains a major challenge, limiting efforts to rationally design and control microbial communities. Because current holistic (meta-omics) and reductionist (isolation- or single-cell-based) approaches struggle to capture these emergent microbial community functions, this study explores an intermediate strategy: analyzing simple sub-community combinations to enable a bottom-up understanding of community-level functions. To examine the validity of this approach, we used a nine-member synthetic microbial community capable of degrading the environmental pollutant aniline, and systematically generated a dataset of 256 sub-community combinations and their associated functions. Analyses using random forest models revealed that the sub-community combinations of just three to four species enabled the quantitative prediction of functions in larger communities (5–9 member; Pearson’s r = 0.780–0.801). Prediction performance remained robust even with limited sub-community data, suggesting applicability to more diverse microbial communities where exhaustive sub-community observation is infeasible. Moreover, interpreting models trained on these simple sub-community combinations enabled the identification of key species and interspecies interactions that strongly influence the overall community function. These findings provide a methodological framework for mechanistically dissecting complex microbial community functions through sub-community-based analysis.