|
Abstract:
|
The human microbiome consists of all the micro-organisms in a community. Due to its central role in many aspects of human health, it is important to understand the dynamic relationships between the organisms and how such dependence structures change in disease state. However, excessive zeros and compositional nature of the data from metagenomics sequencing make it challenging to learn the dependence structure among the bacteria. We present several Copula models to model such dependence, with a mixture of zero and beta density as the marginal densities. The resulting dependency parameters have biologically relevant interpretation. In addition, both zero component and non-zero proportions of the data contribute to the estimate of the dependence parameters. Our work shows that a two-stage estimation method gives valid inference of the parameters in the models. We illustrate the methods using simulations and analysis of several real microbiome data sets.
|
