Abstract:
|
The biological interactions among the microbial taxa are critical in understanding the underlying role microorganisms play in natural environment and human systems. To reveal microbial interaction network, statistical methods that rely on graphical models have been developed to find out the conditional dependence structure among the microbial taxa. Due to the compositional nature, the microbiome data are often transformed before modeled via graphical models, such as the additive log-ratio (alr) transformation used in COMP-GLASSO. However, it is still an open question whether the true and estimated interaction network is robust to the choice of the reference taxon in the alr transformation. In this paper, we first establish the reference- invariance property of a subnetwork of interest based on the alr transformed data with respect to different choices of reference. Then, we propose a reference-invariant version of COMP-GLASSO by modifying the penalty in its objective function, penalizing only the invariant subnetwork. We will illustrate the reference-invariance property of the proposed method under a variety of simulation scenarios and applications to real microbiome studies.
|