Table 1.
Differential abundance tools compared in this study.
| Tool (version) | Input | Norm. | Trans. | Distribution | Covariates | Random effects | Hypothesis test | FDR Corr. | CoDa | Dev. For |
|---|---|---|---|---|---|---|---|---|---|---|
| ALDEx2 (1.18.0) | Counts | None | CLR | Dirichlet-multinomial | Yes* | No | Wilcoxon rank-sum | Yes | Yes | RNA-seq, 16S, MGS |
| ANCOM-II (2.1) | Counts | None | ALR | Non-parametric | Yes | Yes | Wilcoxon rank-sum | Yes | Yes | MGS |
| Corncob (0.1.0) | Counts | None | None | Beta-binomial | Yes | No | Wald (default) | Yes | No | 16S, MGS |
| DESeq2 (1.26.0) | Counts | Modified RLE (default is RLE) | None | Negative binomial | Yes | No | Wald (default) | Yes | No | RNA-seq, 16S, MGS |
| edgeR (3.28.1) | Counts | RLE (default is TMM) | None | Negative binomial | Yes* | No | Exact | Yes | No | RNA-seq |
| LEFse | Rarefied Counts | TSS | None | Non-parametric | Subclass factor only | No | Kruskal–Wallis | No | No | 16S, MGS |
| MaAsLin2 (1.0.0) | Counts | TSS | AST (default is log) | Normal (default) | Yes | Yes | Wald | Yes | No | MGS |
| MaAsLin2 (rare) (1.0.0) | Rarefied counts | TSS | AST (default is log) | Normal (default) | Yes | Yes | Wald | Yes | No | MGS |
| metagenomeSeq (1.28.2) | Counts | CSS | Log | Zero-inflated (log-) Normal | Yes | No | Moderated t | Yes | No | 16S. MGS |
| limma voom (TMM) (3.42.2) | Counts | TMM | Log; Precision weighting | Normal (default) | Yes | Yes | Moderated t | Yes | No | RNA-seq |
| limma voom (TMMwsp) (3.42.2) | Counts | TMMwsp | Log; Precision weighting | Normal (default) | Yes | Yes | Moderated t | Yes | No | RNA-seq |
| t-test (rare) | Rarefied Counts | None | None | Normal | No | No | Welch’s t-test | Yes | No | N/A |
| Wilcoxon (CLR) | CLR abundances | None | CLR | Non-parametric | No | No | Wilcoxon rank-sum | Yes | Yes | N/A |
| Wilcoxon (rare) | Rarefied counts | None | None | Non-parametric | No | No | Wilcoxon rank-sum | Yes | No | N/A |
*The tool supports additional covariates if they are provided. ANCOM-II automatically performs ANOVA in this case, ALDEx2 requires that users select the test, and edgeR requires use of a different function (glmFit or glmQLFit instead of exactTest).
ALR additive log-ratio, AST arcsine square-root transformation, CLR centered log-ratio, CoDa compositional data analysis, CSS cumulative sum scaling, FDR Corr. false-discovery rate correction, MGS metagenomic sequencing, RLE relative log expression, TMM trimmed mean of M-values, Trans. transformation, TSS total sum scaling.