Mixed models for correlated compositional data: applied to microbiome studies in Indonesia

Abstract

The main objective of this thesis is to improve the

understanding of the role of helminth infections on the human gut microbiome

and consequently on the human immune responses. We utilized the observations

from a household randomized controlled trial in a helminth endemic area in Indonesia, where the microbiome data is the main outcome of interest. The microbiome data obtained via high throughput sequencing possess a compositional structure. Therefore, each bacteria category correlates with each other. In addition, the repeated measurement of the microbiome data imposes an extra correlation within bacteria category over time. We first developed a new statistical methodology to model the effect of infection as well as deworming on microbiome composition by taking into account the compositional nature of microbiome data as well as the correlation due to repeated measurements. Secondly, to model the effect of helminth infection on the gut microbiome composition and immune responses, a joint model is developed. This model is able to estimate the correlation between microbiome composition and immune responses.