Human gut microbial communities perform important roles including digesting nutrients, resisting pathogens, and training the immune system. Research interest has therefore focused on identifying bacterial taxa that carry out specific functions and their dynamics. Yet, statistical and experimental challenges hinder this endeavor. The relative nature of microbiota sequencing-based surveys induce artifacts when conventional statistical analyses are performed. Human and animal digestive tracts are difficult to frequently sample and hard to finely manipulate.
Here, we describe two tools we have developed to identify microbes with key ecological roles and analyze their dynamics. The first tool is a statistical method that uses Bayesian state space models to create interpretable models of which bacteria respond to a given perturbation or drive a measurable community function. Crucially, this model is built upon a microbiota data transform we have developed (PhILR), explicitly designed to mitigate artifacts associated with relative abundance sequencing data. The second tool we introduce is an ex vivo human gut model, in which we can culture hundreds of gut bacterial species for weeks. This tool allows us to manipulate and sample human gut microbiota with arbitrary frequency. We combine these tools to reveal how the abundance and metabolism of specifc gut taxa change within hours of nutrient perturbation. Moreover, we describe how these dynamics have “memory,” meaning that microbial responses depend on how many perturbations have previously been carried out.