TUESDAY Seminar - Emily Miraldi!

May 20, 2014

We have a special seminar this coming Tuesday, May 20 in a new room! Emily Miraldi, a CSB Alum, will be speaking to us about her research - Accurate inference of microbial ecological networks. Emily is a former grad student in Forest White's lab, and is now a post doc in Rich Bonneau's lab at NYU. Her abstract is below.


Tuesday, May 20, 12:30

Grier A room, Building 34-401A



Accurate inference of microbial ecological networks

(In collaboration with Zach Kurtz, Christian Mueller, and Rich Bonneau)


16S-ribosomal sequencing techniques provide snapshots of microbial communities, enabling quantification of relative microbial abundances across diverse ecosystems.  While changes in microbial community structure are demonstrably associated with certain environment conditions (from metabolic and immunological health in mammals to ecological stability in soils and oceans), identification of the underlying microbial mechanisms require new statistical tools, as microbiome abundance data presents several challenges.  Foremost, the abundances of microbial operational taxonomic units (OTUs) from 16Sr datasets are compositional. That is, each OTU's counts are normalized to the total number of counts in the sample. Thus, microbial abundances are not independent, and traditional statistical metrics (e.g., correlation) for the detection of OTU-OTU relationships can lead to spurious results.  Equally important, microbial sequencing-based studies typically measure hundreds of OTUs on only tens to hundreds of samples. This suggests that any abundance-based model of microbial interactions is likely under-powered/under-determined and additional information (or assumptions) are required to infer accurate networks. 

To this end, we developed a statistical method for the inference of ecological interactions from 16Sr sequencing datasets that addresses both of these issues. We integrate data transformations developed for compositional data analysis with sparse estimation of the ecological network interactions using a Gaussian graphical model framework.  Because no large-scale microbial ecological networks have been experimentally validated, we tested our method on representative synthetic network types (e.g., cluster, scale-free), supposing that any method that successfully infers the sets of interactions underlying these extreme networks from synthetic data will also perform well in the context of true microbiome datasets, whose underlying ecological topology is likely to be some mixture of these extreme network types.  We generated correlated microbial count data by sampling from a zero-inflated negative binomial distribution, which enabled us to fit OTU distributions from experimental datasets for given network topologies. We compare our method to previously published tools and demonstrate high-quality network recovery. Further, we use our methods to infer specific microbial taxon-taxon interactions and ecological network structure in the uncharacterized gut and other microbial ecosystems.