cfaed Seminar Series

Abstract 

In recent years, 16S-rRNA and other environmental sequencing efforts have revealed phylogeny diversity and abundances of microbial populations across many ecosystems. An important step toward understanding microbial communities on a systems level is the robust identification of association patterns among microbes from such high-throughput data. In this talk I outline how state-of-the-art high-dimensional statistics methods combined with compositional data transformations can robustly and reproducibly infer microbial association networks from microbial sequencing data. Our computational framework, termed SPIEC-EASI (pronounced speak-easy), can learn various parametric, non-parametric, and latent graphical models of microbial associations, and includes novel generalized stability-based approaches for robust network selection.
Using this framework, we inferred association networks from a large number of publicly available data sets across many ecosystems, ranging from gut and freshwater habitats to urban environments. Statistical analysis of the resulting microbial association networks revealed general topological features within and across habitats that pertain to the ecological stability of microbial communities. These findings shed new light on the organization principles of microbial communities and present a promising step toward the unification of ecological theory and data-driven systems microbiology