For many years, our scientific knowledge about microbes was limited to those few species that could be cultured in the lab. With the advent of high throughput sequencing methods, scientists started exploring the genomic information of microbial communities, obtaining a list of microbial identities and overall potential genes present in any ecosystem. Microbial identities are determined by sequencing a marker gene of interest (amplicon) and grouping sequences into Organizational Taxonomic Units (OTUs) where each OTU is believed to come from the same species or genera. The gene potential is established by sequencing all genomes (shotgun sequencing) present in the ecosystem. Naturally, identifying what microbial species and genes are present is just the beginning as scientists seek to understand the interactions among those species and how those genes are expressed in fulfilling biochemical exchanges. Recently, microbial ecologists have begun using microbial association network inference to describe the non-random organization of microbial communities and the emerging ecosystem dynamics. Network inference uses the statistical and topological properties of a set of entities (nodes) and the interconnections between them (edges) to infer the nature and strength of interactions. Two approaches that use network interference at their core are microbial co-occurrence networks and metabolic network models. The first uses high-level data to examine shifts in microbial communities across environmental variables in order to understand the processes driving microbial community structure and dynamics. The latter uses a mechanistic understanding of biochemistry and cellular growth to generate predictive models and explore the potential metabolite exchanges among species. However, integrating metabolic models into microbial ecology workflows is happening slowly primarily because the models require fully annotated genomes as input data, which is generally only obtained from the most expensive sequencing technologies (shotgun). Here, I propose a new method (probabilistic OTU modeling) to generate metabolic models from inferred full genomes using amplicon sequences as data inputs. The methodology relies on mapping the amplicon sequence to a reference database and identifying a pangenome that is phylogenetically related to that sequence. Then, the superset of genes present in the pangenome is used to build the corresponding metabolic model. Community models can be created as a simple aggregation of individual metabolic models. In this study, I created amplicon-based co-occurrence networks to characterize the dolphins and environmental microbiomes at the Shedd Aquarium’s oceanarium habitat. I explored how each microbiome potentially influences other microbiomes and how the addition of probiotics to the dolphin’s diet can affect their skin, chuff, and gut microbiomes. In a different built environment study, I similarly applied amplicon and metabolite-based co-occurrence networks to study the microbe-metabolite succession on common construction woods subject to high relative humidity. I applied the new probabilistic OTU modeling method to generate metabolic models representing each type of wood and identified biochemical clues that connect microbial taxonomies and signature metabolites to the various types of woods. These findings provide evidence of the usefulness of amplicon co-occurrence analysis combined with community metabolic models to offer concrete evidence of how ecological relationships are established within the built environments.