[Calendar Icon] Events Calendar

Assistant Professor of Biostatistics
Bloomberg School of Public Health
Johns Hopkins University
Baltimore, MD

Evaluating the overall association between the microbial community and a phenotype is usually the first and a pivotal step toward our comprehensive understanding of the complex ecosystem and its health implications. Statistically, kernel machine regression models, such as MiRKAT and its extensions, are among the most popular approaches for such evaluation. MiRKAT proceeds by first constructing a distance (beta-diversity) matrix, transforming the distance matrix into a kernel similarity matrix, and assessing the association by comparing whether samples with similar microbiome profiles also have similar phenotype values. Recently, large epidemiological studies increasingly employed more-complex study designs for microbiome research, such as family-based or longitudinal studies, and/or collected other types of high dimensional structured data, such as gene expression. In this talk, we will discuss about two recent extensions of the MiRKAT, for microbiome association testing. aGLMM-MiRKAT is designed to analyze correlated microbiome data, such as in family or longitudinal studies. KRV (kernel RV coefficient) works for evaluation of the association between microbiome and a high-dimensional structured outcome. We will show the applicability of these two approaches both in simulation and real data analysis, and hope that the discussion will generate new research idea and potential collaborations.