Hi
I returned to TCGA raw counts (downloaded from xena) and performed a differential expression analysis between mutation x to "WT", once with DESeq and once with edgeR. I keep default settings. I know the statistics are different (one using geometric the other log-ratio based). However, I got 24k significant results (padj <0.05) with DESeq2 while only about half with edgeR (FDR <0.05). In addition, there are about 7500 common significant genes. The direction (logFC) of the significant genes is similar and almost perfectly matched but not the power.
I noticed that many of the top significant features in the edgeR analysis are related to ribosomal RNA genes, and most of them are marked as "NA" or insignificant in the DESeq2 analysis. There is logic behind those results; it seems DESeq2 is treating them as "noise." Conversely, some results that are expected to be highly upregulated—and indeed are upregulated in DESeq2—show weak statistical strength in edgeR (by logFC)
It's important to note that the comparison is unbalanced (one group has four times as many people as the other group) but both higher than 30**
I would appreciate any suggestion about what should I do further
Best
