Skip to main content

Table 1 The gene ranking performance at different group sizes for all 14 methods

From: Statistical evaluation of methods for identification of differentially abundant genes in comparative metagenomics

AUC0.05 Data set 1: (Qin 2010) Data set 2: (Yatsunenko 2012)
Group Size 3 + 3 6 + 6 10 + 10 3 + 3 6 + 6 10 + 10
edgeR 0.64 0.77 0.85 0.55 0.80 0.92
DESeq2 0.70 0.80 0.86 0.53 0.78 0.90
OGLM 0.64 0.83 0.90 0.44 0.73 0.88
MetagenomeSeq 0.29 0.68 0.87 0.41 0.70 0.86
Metastats 0.19 0.74 0.79 0.18 0.66 0.84
Voom 0.63 0.74 0.78 0.48 0.70 0.83
Sqrt t-test 0.62 0.78 0.85 0.43 0.70 0.86
Log t-test 0.63 0.76 0.79 0.44 0.69 0.84
t-test 0.56 0.77 0.85 0.38 0.67 0.84
Welch t-test 0.47 0.74 0.83 0.32 0.62 0.82
WMW ----- 0.75 0.83 ----- 0.66 0.84
Binomial 0.40 0.46 0.48 0.33 0.51 0.67
GLM 0.32 0.35 0.36 0.34 0.52 0.68
Fisher’s exact test 0.32 0.35 0.36 0.33 0.51 0.67
  1. Each listed value is the normalized area under curve up until a false positive rate of 0.05. Higher values represent higher gene ranking performance. The results are calculated based on 100 resampled metagenomes. The Wilcoxon-Mann–Whitney test was not evaluated at the smallest sample size (3 + 3) due to lack of samples. The full area under curve (AUC) measurements are available in Additional file 1: Table S1