STATISTICAL APPROACHES FOR MATCHING THE COMPONENTS OF COMPLEX MICROBIAL COMMUNITIES

Abstract

A hierarchical Bayesian model called BiomeNet can be used to identify the functional units of metabolic reactions, subnetworks and community-level metabolic networks. The framework models metabolic structures by assuming each sample consists of many tightly connected subnetworks, which in turn are comprised of di fferent reactions. When applying the method the number of subnetworks L must be pre-specified. When L is set larger in BiomeNet, the inferred structures of the subnetworks are expected to com out in a more trivial form. Three methods, LASSO, NNLS and a new method, MJSD, are applied to match a subnetwork in one analysis (say, when L=100) with several subnetworks when L is increased (say, when L=200). RSS and JSD are applied a matching criteria to conduct multiple tests to judge the significance of the matches. From the results, I am able to identify those "predominant" subnetworks and give a reasonable conjecture that those "predominant" subnetworks always come out as unbroken blocks for any larger L values.