A multivariate zero-inflated logistic model for microbiome relative abundance data

摘要

The human microbiome plays critical roles in human health and has been linkedto many diseases. While advanced sequencing technologies can characterize thecomposition of the microbiome in unprecedented detail, it remains challengingto disentangle the complex interplay between human microbiome and disease riskfactors due to the complicated nature of microbiome data. Excessive number ofzero values, high dimensionality, the hierarchical phylogenetic tree andcompositional structure are compounded and consequently make existing methodsinadequate to appropriately address these issues. We propose a multivariatetwo-part model, zero-inflated logistic normal (ZILN) model to analyze theassociation of disease risk factors with individual microbial taxa and overallmicrobial community composition. This approach can naturally handle excessivenumbers of zeros and the compositional data structure with the zero part andthe logistic-normal part of the model. For parameter estimation, an estimatingequations approach is employed and enables us to address the complex inter-taxacorrelation structure induced by the hierarchical phylogenetic tree structureand the compositional data structure. This model is able to incorporatestandard regularization approaches to deal with high dimensionality. Simulationshows that our model outperforms existing methods. Performance of our approachis also demonstrated through the application of the model in a real data set.