A1294
Title: Microbial co-abundance network with community detection
Authors: Yan Li - Auburn University (United States) [presenting]
Abstract: Microbial network analysis holds the potential to decipher complicated ecological interactions between microbes and, therefore, enhance the understanding of microbial functionality. However, existing methods are inadequate to accommodate the compositionality of microbiome data or extract concise and insightful dependence structures from high-dimensional data. A new network framework with structural constraints to the microbiome data analysis is proposed based on the notion of Laplacian-constrained Gaussian graphical models. The proposed Laplacian constraint model provides a perfect fit for the unique features of centered log-ratio transformed compositional data with a linear constraint. Under the proposed framework, zero-sparsity and nuclear-norm regularization methods and an efficient computation algorithm are developed to achieve a highly interpretable co-abundance network with biologically meaningful community separation. Theoretical properties are explored for the proposed model. The efficacy of the proposed methods is demonstrated in extensive simulation studies and real gut microbiome studies.
