B1935
Title: Compositional differential abundance analysis for health-microbiome associations and controlling false discoveries
Authors: Siyuan Ma - Vanderbilt University Medical Center (United States) [presenting]
Abstract: The ubiquitous differential abundance (DA) analysis for the microbiome examines each microbe as isolated from the rest of the microbiome by design. This does not properly account for the microbiome's compositional nature or microbe-microbe ecological interactions and can lead to confounded, false discovery findings. To remedy these issues, compositional differential abundance (CompDA) analysis is presented, a novel approach to health-microbiome association. CompDA identifies health-related microbes by examining the microbiome holistically, which a) accounts for the data compositionality and ecological interactions, and b) has clear interpretations corresponding to host health as affected by microbiome-based interventions. Methodology-wise, CompDA implements recent advances in high-dimensional statistics. It can be flexibly adapted to many common tasks in modern microbiome epidemiology, including enhancing microbiome-based machine learning by providing rigorous p-values to prioritize important features. The performance of CompDA is validated and compared against canonical microbiome association methods including DA with extensive, real-data-informed simulation studies. Lastly, novel and consistent findings of CompDA are reported in application studies, including a) recently reported microbial signatures of colorectal cancer from cross-study machine learning, and b) well-established microbial associations of early-onset Crohn's disease in a pediatric cohort.
