Latent Structure and Causal Mediation Models for Human Microbiome Sequencing Data

The exploration of human microbiome sequencing data has emerged as a popular research area in recent medical science. Nonetheless, owing to inherent characteristics of microbiome data such as excess zeros, high skewness, sparsity, and a profusion of outliers, the analytical process faces some substantial hurdles. Consequently, the overarching aim of this thesis is to develop statistical methodologies that enable comprehension of human microbiome sequencing data analysis, encompassing investigation of latent structures and detection of potentially mediating roles played by certain microbial taxa between exposure/treatment to the interest outcomes. The first objective is to develop clustering algorithms which can be employed for microbiome datasets, seeking to unveil underlying data structure and potential subcategories. A novel approach, employing a customizable distance metric, is developed to eliminate the presence-absence bias linked with sparse count data while accurately measuring sample distances.Furthermore, when analyzing the role of microbial taxa as mediators - linking exposure to a study outcome via causal mediation analysis of microbiomes - single or multiple microbial entities may emerge as mediators. My second objective is to construct a causal mediation framework utilizing a weighting-based model specifically designed for a single zero-inflated microbial, demonstrating that our proposed framework precisely quantifies the mediation effect of specific taxa with minimum bias. Lastly, managing high-dimensional zero-inflated microbiome mediators within a counterfactual framework becomes crucial given that these mediators are correlated and often present with zero inflation. My third objective is to develop a method that incorporates the aforementioned distance metric in Objective One, specifically employed for microbiome data, capable of solidly assessing the frequency of zeros in the dataset, therefore tackling the dimensionality reduction faced by the microbiome data to prevent overfitting during mediation analysis. All approaches are subjected to comprehensive theoretical deduction and simulation evaluation. Concurrently, I demonstrate the model performance through real data applications. A gut microbiome dataset of Parkinson's disease was used for the first objective. The implementation of the second and third objectives utilizes the DIABIMMUNE Study to examine the mediation effect of breastfeeding on allergic reactions of infants through their gut microbiome.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.