Question: RNA-seq Normalisation - normalise all samples in experiment or only the ones used in the comparison?
gravatar for juls
20 months ago by
juls0 wrote:

Hi all,

I have a general question about normalisation.

A collaborating group has performed RNA-seq on 15 samples - three different groups, 5 samples each. Now they realised that they are only interested in a two group comparison though and they told me to forget about the third group. My question now is the following: should I normalise (I am using TMM) only the samples from the 2 groups I want to compare or still all of them together since they were measured together and belonged to the original experimental design? 

I am grateful for advice. 

Thank you!

ADD COMMENTlink modified 20 months ago by Gordon Smyth35k • written 20 months ago by juls0
gravatar for Ryan C. Thompson
20 months ago by
The Scripps Research Institute, La Jolla, CA
Ryan C. Thompson6.9k wrote:

Leaving in the 3rd group should make barely any difference in the normalization step. In the absence of large global changes in expression, normalization relationships between samples should be approximately transitive, so adding a third group should not substantially change the relative normalization of the samples in the first 2 groups. You can verify that this is the case by computing the normalization factors both ways and then comparing them. (Remember that the absolute values may change, but the relative ratios should not change much.)

In general, it is recommended to use all the available samples from an experiment in the analysis, even if you are not interested in differential expression for some of those samples. The main reason for this is that having more samples allows more accurate estimation of the gene dispersion values.

ADD COMMENTlink written 20 months ago by Ryan C. Thompson6.9k

Dear Ryan, 

Thank you very much for your quick reply. As the resulting DE gene list indeed changes, I wasn't sure which way is correct. 





ADD REPLYlink written 20 months ago by juls0

Hi Ryan,

I have a follow-up question. Since you mentioned "to use all the available samples from an experiment in the analysis"  - would you also include them in the voom transformation/design matrix then?






ADD REPLYlink modified 19 months ago • written 19 months ago by juls0

Yes, if you don't normalize and run voom on all the samples together, then there's no way to analyze them together.

ADD REPLYlink written 19 months ago by Ryan C. Thompson6.9k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 330 users visited in the last hour