Abstract

Human microbiome is a diverse and dynamic ecosystem that plays a fundamental role in normal human physiology and health. Thousands of distinct microbial strains inhabit our bodies, with remarkable individual variation. Large-scale population studies provide a window to the socio-demographic drivers of microbiome composition and their associations with prevalent and incident disease occurrence. The statistical properties of microbiome data, such as sparsity, zero-inflation, compositionality, collinearity, high dimensionality, and hierarchical structures, are representing challenges for statistical modeling, however. This talk will discuss recent advances in statistical modeling of hierarchical multi-assay data arising from microbiome profiling studies across multiple observation layers ranging from the taxonomic composition of the community to its functional aspects and interactions with the human host. This talk will provide an overview of important statistical considerations in microbiome research, and assess the potential of microbiome profiling in predicting the future mortality and disease risk in prospective population cohort studies.