Normalization factor vs. scale factor

[yuhuihui2011]

Hello,

The SF in the output file (test_SF.txt) is scale factor and DGEList function in edgeR needs normalization factors. So I think the
norm.factors should be 1/SF.

dge <- DGEList(counts = counts, group = GROUP, norm.factors = SF)

should be:

dge <- DGEList(counts = counts, group = GROUP, norm.factors = 1/SF)

And in the part: What happens if 100% complete loss is expected? You just used 1/SF. This is a little confusing.

Would you please clarify this issue?

Thanks

[hongjianjin]

It is always correct to use SF in this way
dge <- DGEList(counts = counts, group = GROUP, norm.factors = SF)

When 100% complete loss is true, please note that there is NO accurate way to calculate a SF because there is no true signal in these samples. The proposed transformation is to assign an arbitrary value 1 to this sample group and scale down the other group to make a pseudo- scaling factors. It is ONLY suitable for a naïve design having only two groups. In a design like controls (n=3, real SFs= 2, 2, 3), vs ko(n=3, real SFs=NA, NA,NA) . The pseudo-SFs will be 0.5,0.5,0.3 (controls) and 1,1,1 (ko) so that you can see a global decrease in ko group.