Statistical Challenges for Human Microbiome Analysis

DNA sequencing technologies have revolutionized microbiome studies. In this work we analyze microbiome data from an HIV study focused on the characterization of microbiome composition in HIV-1 infected patients. A 155 cohort of HIV infected and non-infected individuals is analyzed to characterize dietary and gut microbiome association in this group of patients. A penalized Dirichlet Multinomial regression model has been considered. The assumed underlying Dirichlet distribution in this modelization provides additional flexibility to the multinomial model which results in a better fit of the typically overdispersed microbiome data.