Hi all,
For a bit more background (and for the sake of posterity), Gordon has
previously discussed the rationale of tunable (or not) prior.counts in
edgeR::cpm and voom in this support forum thread:
https://support.bioconductor.org/p/59846/#59917
-steve
On Thu, Feb 16, 2017 at 8:00 PM, Dario St
Good day,
Now I notice the differences in how the prior counts are applied. In edgeR's
cpm:
prior.count.scaled <- lib.size/mean(lib.size) * prior.count
lib.size <- lib.size + 2 * prior.count.scaled
......
t(x) + prior.count.scaled
but in limma's voom:
t(counts + 0.5
They wouldn't be exactly consistent even if they used the same prior count,
since the calculations are not identical. edgeR normalizes the prior count
by each library's normalization factor so that log fold changes are always
squeezed toward zero, while voom, if I understand correctly, does not
nor
Good day,
The cpm function in edgeR uses a default offset of 0.25 and voom in limma uses
0.5 (and provides no user modification) to calculate the base 2 logarithm of
the counts per million. Might these be made consistent?
--
Dario Strbenac
University of Sydne
