Phenomenological modeling reveals the emergent simplicity of host-associated microbiomes.

Correlated variation between host phenotypes and microbiomes suggest that emergent global variables may simultaneously describe the microbial ecosystem and the host. Mechanistic models cannot yet identify these descriptors because of their inherent complexity. To that end, we present a phenomenological model based on the consumer/resource framework wherein microbial species and hosts' phenotypes are coupled through their shared dependence on a small number of generalized resources (latent variables). We show that animal microbiomes are surprisingly low-dimensional; the number of latent variables needed to accurately describe these ecosystems is significantly smaller than the typical number of resources or microorganisms present. The model reproduces key metrics of biodiversity through probabilistic sampling of the latent variables. It also identifies host phenotypes that significantly determine the latent space, and therefore predict the microbiome composition from host phenotypes, and vice versa. Finally, clustering of hosts using their latent representation identifies subsets of hosts with context-specific phenotype-microbiome associations. We believe that this phenomenological approach will be key to quantitative exploration of host-associated microbiomes.