Simulation studies to assess the power of set testing methods for microbiome data
With advances in sequencing methods, the study of the microbiome has greatly increased. Microbiome data, in the form of an OTU or ASV count table, can be used to identify specific ASVs that function differently across treatment conditions. Such analysis is deemed differential abundance analysis. ASVs are grouped by their taxonomic rank, and ASVs sharing the same rank have similar biological traits. By studying groups or sets of ASVs, and identifying if the set is differentially abundant, the biological interpretation of a microbiome study is enhanced. We review current approaches in set testing methods and apply them to a microbiome data set from a 2017 study. We propose a new set testing method based on an existing Poisson hurdle model, and compare performance across all methods through a model based simulation study. We find that under certain conditions, our proposed model outperforms existing approaches. We discuss the limitations of our model and conclude that more simulation studies, specifically non-parametric simulation studies, are needed to better compare across possible methods.