Shift in expression of found reference-genes in voom normalized counts

[tycho-kirchner]

Hi,
and thanks for sharing this package. I ran moose on the public dataset GSE102481 where I assume a global transcriptional shutdown. Indeed, while DESeq reports 4534 up- and 3458 downregulated genes, moose-limma-voom finds only 1367 up- but 10365 downregulated genes. I used a limma-voom pipeline based on your script and the moose normalized count distribution of the reference genes looks promising. However, writing out the voom normalized counts, we see in the boxplot and density plot, that voom apparently scaled the counts again, leading to a distortion of the distribution of reference genes. I suspect that voom, performing its calculations on CPM, does not appropriately respect moose's scaling factors.
Am I correct?

Thanks
Tycho

PS: I attached raw counts, moose normalized counts, voom normalized counts and my limma-script. Usage:
dgea-limma.R --counts normalized.out -o "$PWD"

data.zip