Figure 9 : In the Platinum Spike dataset, all normalization methods resulted in similar detection of differential gene expression, with MedianCD and SVCD normalization being only marginally better.

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

Figure 9

Panels display ROC curves, with the true positive rate versus the false positive rate (a), or the number of true positives versus the number of false positives (b). Each curve shows the results obtained after applying the four normalization methods plus Cyclic Loess normalization (same colors and symbols as in Figs 3, 6, 7 and 8; black curve with empty black circles, Median normalization; red curve with empty red up triangles, Quantile normalization; green curve with filled green circles, MedianCD normalization; blue curve with filled blue up triangles, SVCD normalization; magenta curve with filled magenta diamonds, Cyclic Loess normalization). Dashed curves with lightly filled symbols show results when the list of known negatives was provided to MedianCD, SVCD, and Cyclic Loess normalization. As in Fig. 8, the two points per normalization method show results when controlling the false discovery rate (FDR) to be below 0.01 (left point) or 0.05 (right point). Dashed lines in (b) show references for actual FDR equal to 0.01, 0.05, 0.1, 0.2, or 0.5 (from left to right). Compared to the Golden Spike dataset (Fig. 8), the difference between normalization methods in the resulting degradation of the FDR was smaller for this dataset.