Skip to main content
Figure 5 | Genome Biology

Figure 5

From: Statistical methods for ranking differentially expressed genes

Figure 5

Spike data. (a) Spike data with 14 DEGs. Three arrays from each group in the Affymetrix spike experiment were used. The granularity of Wilcoxon shows up here as a lack of performance, and at this sample size Wilcoxon is not an option. SAM starts off optimistically, and then falls back when the lists become longer and the false positives more, and samroc and Bayes catch up with it. In particular, the latter performs strongly. (b) Spike data with 154 DEGs. The spike data with an added 140 changed genes obtained from adding permuted residuals to group means for the 14 spiked genes, generating three arrays per group. This makes the percentage of DEGs just above 1%. We see that samroc improves considerably compared to (a), and now shows the best performance for a wide range of top list sizes. (c) Spike data with 714 DEGs. The spike data with an added 700 changed genes obtained from adding permuted residuals to group means for the 14 spiked genes, generating three arrays per group. This makes the percentage DEGs just above 5%. Now samroc takes the lead, and when the false positives reach roughly 10%, it is passed by the Bayes method. This point corresponds to a top list of roughly 1,400.

Back to article page