Network Analysis for Metagenomic Compositional Data

• Activity Number: 366
• Type: Invited
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: General Methodology
• Author(s): Hongzhe Li*+ and Yuanpei Cao and Wei Lin
• Companies: University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
• Keywords: sparse covariance ; log-ratio ; low-rank projection ; microbiome

Abstract

Metagenomics and high-throughput sequencing have enabled the study of ecosystem structure and dynamics of microbes to great depth. Co-occurrence and correlation patterns found in these data sets can used for the prediction of species interactions in environments ranging from human guts to skins. However, the data from such metagenomic studies are of high dimensional compositional nature. Standard use of correlations to compositional data can lead to false associations among the microbes. In this paper, we propose an estimation method for sparse covariance matrix for compositional data. Our method is based on decomposing the Aitchison's variation matrix T as sum of rank-2 matrix and a sparse base covariance matrix. We provide theoretical results on estimation errors and sparse recovery. We illustrate the method using a metagenomic study of obesity and Crohn's disease.