|
Abstract:
|
The microbiome plays an important role in human health. Limited to the intrinsic characteristics of microbiome data, the appropriate statistical methodology for detecting differential microbes is highly demanded. Here, we introduced a novel statistical framework called Logarithm Ratio Tree Test (LRTT) by incorporating phylogeny to do differential abundance (DA) analysis. We first extend Zero-inflated Generalized Dirichlet-multinomial (ZIGDM) to Zero-inflated Dirichlet-Tree- Multinomial (ZIDTM) distribution. It can describe the sparsity of microbiome data well while with a more flexible covariance structure. To increase statistical power, we use the likelihood ratio test to choose best-fitted distribution and the corresponding posterior mean to smooth zero. We also use the maximum posterior mean of the non-zero count to detect outliers. Then with the preprocessed data, LRTT takes microbes relationship into account which described by a phylogenetic tree to build log-ratio transformation adaptively. Through extensive simulations and two real data sets, we demonstrate that our smooth strategy and LRTT framework are both superior to existing methods.
|
