Statistics Seminar Christian Müller, 5/14/14 @ 4:15 PM

Event Layout

Wednesday May 14 2014

Statistics Seminar Christian Müller, 5/14/14 @ 4:15 PM

4:15pm @ G01 Biotechnology

This week's Statistics Seminar Speaker will be Christian Müller from New York University.

Title: Inference of Microbial Ecological Interaction Networks with SPIEC-EASI

Abstract: In recent years, high-throughput sequencing techniques have enabled quantification of microbial communities across diverse ecosystems. One of the key goals of these research efforts is to decipher the underlying ecological interactions among the microbial species.  The inference of such networks requires new tools as microbiome abundance data present several challenges. For instance, the abundances of microbial Operational Taxonomic Units (OTUs) are usually compositional because they are normalized to the total observed counts. In addition, in present sequencing-based studies the number of identified species p largely exceeds the number of available samples n. Any abundance-based inference method thus operates in the underdetermined regime (p>>n), and additional structural assumptions about the underlying network are required for accurate inference.

We here introduce SPIEC-EASI (SParse InversE Covariance selection for Ecologial ASsociation Inference), a novel computational pipeline for estimating sparse ecological networks using robust proportionality measures of microbial compositions. SPIEC-EASI first applies log-ratio transforms to the compositional data set and then infers networks by solving the sparse inverse covariance selection problem (or approximation of that) using tools from high-dimensional statistics. In addition, the SPIEC-EASI package comprises several tools to construct realistic synthetic data sets (i) that share many properties of publically available data and (ii) that are generated from a wide range of underlying network topologies. We present different benchmark scenarios that show the superior recovery performance of our method over competing state-of-the-art methods in the field. We also apply SPIEC-EASI to recent microbiome datasets (such as the American Gut project) and characterize the topology of the inferred community networks.

Joint work with Zachary Kurtz, Emily Miraldi, Rich Bonneau, Eric Alm, and Martin Blaser.

Refreshments will be served after the seminar in 1181 Comstock Hall.