Beta-Diversity Discriminatory Power: Comparison of PERMANOVA, Mirkat, and Using Standard Microbiome Reference Groups

• Activity Number: 102 - Challenges and Developments in Microbiome Data Science
• Type: Invited
• Date/Time: Monday, July 29, 2019 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Mitchell Henry Gail* and Yunhu Wan
• Companies: National Cancer Institute, Division of Cancer Epidemiology and Genetics and National Cancer Institute, Division of Cancer Epidemiology and Genetics
• Keywords: microbiome; beta-diversity; power; permanova; Mirkat; standard reference method

Abstract

Comparisons of microbiome communities are often based on pairwise beta-diversity measures, such as Bray-Curtis. For two-group discrimination, a permutation test (PERMANOVA) and a kernel expansion test (Mirkat) are used. A recent proposal computes mean distance from a microbiome test sample to a set of standard reference samples. If several different types of reference samples are used, each test sample has a vector of mean distances. Hotelling’s T-squared test, can be used on these vectors. We compared the discriminatory power of PERMANOVA, Mirkat and use of standard reference groups. We used 16S data from the Human Microbiome Project for the groups to be discriminated (e.g. skin versus saliva samples). The two groups are easy to discriminate, but by using mixtures, we created a sequence of increasingly difficult discrimination problems. Simulations determined power. Although each method showed good discriminatory power, preliminary data indicate that the reference group method performs as well or better than the more complex methods for Bray-Curtis dissimilarity. Additional studies are needed to compare power for other beta-diversity measures.