Table 2.

Differential abundance testing methods

Method	Description
t-test	Welch two sample t-test. We use the R built-in t.test function with default parameters. This test applies to either raw counts or transformed data.
Wilcoxon	Wilcoxon rank-sum test. We use the R built-in function wilcox.test with default parameters. This test applies to either raw counts or transformed data.
DESeq2	A popular method from the field of RNA-seq. It is based on a negative binomial model for raw counts, and is implemented in R package DESeq2 [50]. We use the built-in library size normalization and default parameters.
ANCOM	A novel method for detecting differentially abundant taxa at the ecosystem level using the specimen level relative abundance data. This test is implemented in the R package ancom.R [14]. We use the default setting.
metagenomeSeq	As with ANCOM, this method is developed specifically for microbial datasets. It is based on a zero-inflated Gaussian mixture model for log read counts. We use the function fitFeatureModel in the R package metagenomeSeq [51], with cumulative sum scaling and default parameters.
Wrench	A new technique for compositional bias correction in sparse sequencing count data [20]. It fits a negative binomial log-linear model for reference-based data normalization, and then runs a likelihood ratio test for detecting differentially abundant taxa. We use the functions glmFit and glmLRT in the R package edgeR [52].