Does Gene length corrected TMM [GeTMM] violate any assumptions of TMM normalization?

Question

[O.rka]

I've read and been told that edgeR must take in counts only. I noticed that the RPK values are being fed into the TMM normalization procedure. Is this a correct usage assuming all of the assumptions?

Should this be used for downstream DGE analysis?

Note: I am no expert with these methods but I just wanted to ask the community

https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-018-2246-7

# calculate RPK
rpk <- (x[,2:ncol(x)]/x[,1])
# remove length col in x
x <- x[,-1]
# for normalization purposes, no grouping of samples
group <- c(rep("A",ncol(x)))
#EdgeR
x.norm.edger <- DGEList(counts=x,group=group)
x.norm.edger <- calcNormFactors(x.norm.edger)
norm.counts.edger <- cpm(x.norm.edger)

#GeTMM
rpk.norm <- DGEList(counts=rpk,group=group)
rpk.norm <- calcNormFactors(rpk.norm)
norm.counts.rpk_edger <- cpm(rpk.norm)

# Source:
# https://static-content.springer.com/esm/art%3A10.1186%2Fs12859-018-2246-7/MediaObjects/12859_2018_2246_MOESM4_ESM.docx

[swbarnes2]

For DGE, use raw counts, like the software demands. Other normalizations can be used for things like visualizations.

Answer 1

[Damian Kao]

Technically, RPK values do not violate assumptions of TMM.

TMM is just a technique that tries to find the non-DE portion of the expression distribution by very liberally trimming off outliers. It doesn't matter what kind of expression units you are using.

However, RPK values do violate assumptions for DE analysis. So you cannot use it for downstream DGE.

[salamandra]

I understand your point, but why then in the article they use GeTMM for DE as well? See figure 7 here