Fig. 2

Effect of normalization methods on DGE analysis. a The UpSet diagram shows the intersection size among DGE analyses (at 5% FDR), each using different normalization methods but the same statistical test (moderated t-test from limma package). This result is particularly for the CRC AZA data. All considered normalization methods generally show strong concordance except for quantile normalization (QN). DGE analysis with all normalization methods commonly identified 2820 significantly differentially expressed (SDE) genes, whereas QN resulted in 629 SDE genes that are not shared with other normalization methods. b Distribution of gene-wise coefficients of variation (CV) from the Bottomly data. Each violin plot indicates the quartiles of the distribution (solid horizontal lines). Results based on all six datasets can be found in Additional file 2