Interaction Networks in Microbiome Studies

Abstract

Motivated by the problem of analyzing the microbe–metabolite interaction network in microbiome studies, this article introduces estimators for the high‐dimensional sparse precision matrix of a random vector that consists of both compositional and Gaussian random variables. Based on the idea of centered log‐ratio (CLR) transformation, we construct estimators whose rates of convergence are obtained under various matrix norms as well as an estimator that is consistent for graphical model selection. The proposed method is then applied to analyze the microbe–metabolite interaction network in Crohn's disease patients. We observe that most of the metabolites are associated with very few bacteria, and a given bacterium is associated with a small set of metabolites.

Time-Informed Dimensionality Reduction for Longitudinal Microbiome Studies

“…TEMPTED allows users to choose their own preferred data normalization and transformation. Here, to account for the variation in sequencing depth across samples and the highly skewed distribution of microbiome sequencing data, we apply the centered-log-ratio (CLR) transformation to read counts added by .5 [1,7]…”

“…In recent years, methods designated to model longitudinal microbiome data have been developed. Unsupervised methods like CTF 1 , TCAM 6 and microTENSOR 7 format temporal microbiome data into tabular tensors and apply tensor decomposition to identify low-dimensional structures. Yet, they assume that all hosts have the same sampling time points, which is often unrealistic in clinical settings.…”