Title: Conditional regression based on a multivariate zero-inflated logistic model for human microbiome data
Authors: Zhigang Li - Department of Biostatistics, University of Florida (United States) [presenting]
James OMalley - Geisel School of Medicine at Dartmouth (United States)
Hongzhe Li - University of Pennsylvania (United States)
Abstract: Massive high dimensional human microbiome data is commonly seen in molecular epidemiology research and have substantially increased in complexity to address critical health concerns due to complex data structure. Analysis challenges arise from compositional, phylogenetically hierarchical, sparse and high dimensional structure of microbiome data. Compositional structure could induce spurious relationships due to the linear dependence between compositional components. In addition, the hierarchical structure of microbiome data from the phylogenetic tree generates dependence at the hierarchical levels which poses a further modeling challenge. Furthermore, the sparsity of microbiome data due to excessive zero sequencing reads for microbial taxa remains an unresolved issue in the literature. Coupled with the high dimensional feature, microbiome data raises great challenging problems in the field of mediation data analysis. We will develop a zero-inflated logistic normal model to address these issues. A simulation study will show the performance of the approach and a real study example will be included as well.