Integrated diversity and network analyses reveal drivers of microbiome dynamics.

Microbial communities are key components of ecosystems, where interactions among microbes drive biodiversity and productivity. An increased number of microbiome data sets are available, owing to advances in sequencing; however, standard analyses often focus on community composition, neglecting the complex interactions between co-occurring microbes. To address this, we developed a computational framework integrating compositional and co-occurrence network analyses. We applied this approach to extensive microbial amplicon data sets, focusing on plant microbiota, which typically exhibits high diversity and remains challenging to characterize due to the large number of low-abundance taxa. We show that identifying a subset of representative microbial taxa captures the overall community structure and increases the statistical power. From these taxa, we inferred a large-scale co-occurrence network and clustered microbes with co-varying abundances into units for diversity measurement. This approach not only reduces unexplained variance in diversity assessments but also captures the key microbe-microbe relationships that govern assembly patterns. Furthermore, we introduced a bootstrap- and permutation-based statistical approach to compare microbial networks from diverse conditions. Our method robustly distinguishes meaningful differences and pinpoints specific microbes and features driving those differences. These results highlight the importance of incorporating microbe-microbe interactions in microbiota studies, leading to more accurate and ecologically meaningful insights. Our framework, available as an R package ("mina"), enables researchers to identify condition-specific interactions IMPORTANCE: Understanding microbiome dynamics requires capturing not only changes in microbial composition but also interactions between community members. Traditional approaches frequently overlook microbe-microbe interactions, limiting their ecological interpretation. Here, we introduce a novel computational framework that integrates compositional data with network-based analyses, significantly improving the detection of biologically meaningful patterns in community variation. By applying this framework to a large data set from the plant microbiota, we identify representative groups of interacting microbes driving differences across microhabitats and environmental conditions. Our analysis framework, implemented in an R package "mina," provides robust tools allowing researchers to assess statistical differences between microbial networks and detect condition-specific interactions. Broadly applicable to microbiome data sets, our framework is aimed at enabling advances in our understanding of microbial interactions within complex communities.