Title: Clustering for microbiome data with structural zeros
Authors: Julia Fukuyama - Indiana University (United States) [presenting]
Abstract: Species abundance tables from microbiome studies are famously sparse, with the zeros coming from a combination of stochastic and structural (true) zeros. Both sources of zeros lead to problems in parameter estimation, but they have different sources and need to be modeled differently. In addition, the structural zeros are often strongly associated with covariates, e.g., the identity of the host in human microbiome data. We describe a new method for clustering species in the presence of structural and stochastic zeros. In addition to reducing the dimensionality of the data, this clustering improves interpretability by identifying groups of bacterial species that perform the same functions. The clusters and the corresponding model for structural zeros can be used to examine ecological theories of microbial community assembly and maintenance. We show results on human data and discuss the ecological implications.