Question: Most (90%) DE genes of a contrast were upregulated: implication on the assumptions of TMM-normalisation?
0
gravatar for mikhael.manurung
4 months ago by
Netherlands
mikhael.manurung200 wrote:

Dear all,

If I understand correctly, TMM-normalisation has two assumptions: 1. Most genes are not DE 2. Comparable number of DE genes that are up/down-regulated (i.e. symmetry)

According to my analysis (I am using limma-voom workflow), one of my contrasts yield 200 DEGs, 90% of which were upregulated. Does it mean that the assumptions break?

My second question: is it a good practice to review the previous steps in DE analysis after taking a look at the results?

Best regards, Mikhael

normalization limma edger • 163 views
ADD COMMENTlink modified 4 months ago by Gordon Smyth39k • written 4 months ago by mikhael.manurung200
Answer: Most (90%) DE genes of a contrast were upregulated: implication on the assumptio
0
gravatar for Gordon Smyth
4 months ago by
Gordon Smyth39k
Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia
Gordon Smyth39k wrote:

No, TMM does not make any assumption that there are equal numbers of up and down genes

The edgeR User's Guide says:

TMM is recommended for most RNA-Seq data where the majority (more than half) of the genes are believed not differentially expressed between any pair of the samples.

It doesn't say anything about symmetry.

ADD COMMENTlink modified 4 months ago • written 4 months ago by Gordon Smyth39k

Thank you for the clarification. I got that understanding after reading this paper, particularly the "Normalization by distribution/testing" part.

Best regards, Mikhael

ADD REPLYlink written 4 months ago by mikhael.manurung200

Most normalization methods will have some trouble, not be perfect, when there is a lot of asymmetric DE. But when I say "a lot", I mean as a percentage of the total number of genes, not as a proportion of the DE genes. Your experiment has very little asymmetry. If you have 180 up genes and 20 down genes, then the asymmetry is about 160 genes, which is a tiny percentage of the 10,000 - 20,000 genes that you are probably analysing. This small degree of asymmetry will not cause problems for any normalization method, much less for a very robust method like TMM. The paper you cite agrees that TMM is one of the best choices.

ADD REPLYlink modified 4 months ago • written 4 months ago by Gordon Smyth39k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 229 users visited in the last hour