Skip to main content
Fig. 4 | Genome Biology

Fig. 4

From: Response to "Neglecting normalization impact in semi-synthetic RNA-seq data simulation generates artificial false positives" and "Winsorization greatly reduces false positives by popular differential expression methods when analyzing human population samples"

Fig. 4

Comparison of DESeq2 without or with winsorization, edgeR without or with winsorization, and Wilcoxon rank-sum test on semi-synthetic data generated using the model-based strategy from GTEx data of heart left ventricle vs. atrial appendage. The FDR control (left panel), power given the claimed FDRs (middle panel), and power given the actual FDRs (right panel) under a range of FDR thresholds (i.e., claimed FDRs) from 0.001 to 5%. DESeq2 and edgeR were applied in default (without winsorization) or with three winsorization thresholds (the 93%, 95%, and 97% quantiles) based on the winsorization procedure in Yang et al. [3] . As the true DEGs and non-DEGs were defined without assuming batch effects, DESeq2 and edgeR were applied without normalization. The claimed FDRs, actual FDRs, and power were all calculated as the averages of 30 semi-semisynthetic datasets generated independently

Back to article page

Genome Biology

ISSN: 1474-760X

Contact us