Integrative Analysis of Multiple Microbiome Data Sets: Robust Models Against Biases and Batches

• Activity Number: 144 - Biases, Batch Effects, and Novel Statistical Methodologies: Handling Them in Large-Scale Microbiome Sequencing Studies
• Type: Invited
• Date/Time: Tuesday, August 10, 2021 : 10:00 AM to 11:50 AM
• Sponsor: ENAR
• Author(s): Ni Zhao* and Mengyu He and Runzhe Li
• Companies: Johns Hopkins University and JOHNS HOPKINS UNIVERSITY and Johns Hopkins Bloomberg School of Public Health
• Keywords: integrative analysis; similarity matrix regression; HIV microbiome

Abstract

Integrative analysis of multiple microbiome datasets is necessary to increase the sample size in the effort of identifying consistent signals that are of scientific interest. However, statistically it is challenging because different studies are not directly comparable due to complications such as differential biases and batches. In this talk, we propose robust statistical models for microbiome alpha and beta diversities which are robust to these complications. We propose a two-stage kernel association approach for modeling microbiome alpha diversities that the study-specific characteristics are modeled through a second stage kernel matrix. We further proposed a similarity matrix regression approach for modeling microbiome beta-diversities. All methods are evaluated extensively on simulation data, as well as a real combined datasets that consist the microbiome data from 27 studies on gut dysbiosis in HIV.