Figure 6 : In the synthetic dataset with differential gene expression, and with more than 10% of treatment positives (Fig. 3b), Median and Quantile normalization resulted in less statistical power and uncontrolled false discovery rate.

From: Variation-preserving normalization unveils blind spots in gene expression profiling

Figure 6

The panels display the true positive rate (a) and false discovery rate (b), for each treatment compared to the corresponding control, obtained after applying the four normalization methods (same symbols as in Fig. 3; empty black circles, Median normalization; empty red up triangles, Quantile normalization; filled green circles, MedianCD normalization; filled blue up triangles, SVCD normalization). Both panels have the same order of treatments as in Fig. 3b, i.e. in increasing number of differentially expressed genes identified with SVCD normalization. Differential gene expression was analyzed with R/Bioconductor package limma. The dashed horizontal line in (b) indicates the desired bound on the false discovery rate at 0.05.