Use DESeq to compare paires of conditions and isoforms expression

[Lea Shallev]

Hi,

I have RNAseq of 3 groups (n=2, n=3 and n=3) and I want to calculate the differential expression of 2 ADAR1 isoforms (ADAR p110 and ADAR p150) in group1 vs. group2 and group2 vs. group3.
As I understand (http://seqanswers.com/forums/showthread.php?t=10774 , https://www.biostars.org/p/73491/
and https://stat.ethz.ch/pipermail/bioco...ry/050336.html), I can run DESeq and then simply call nbinomTest for each comparison.

My questitions are:
1. When can I use the normalized table of the 3 conditions to compare pairs of groups and when shouldn't I? If not- which normalization can I do?
2. I use STAR as aligner, and it can open gaps to align spliced reads. Can I use the ratio of different reads at the seperation junction between the 2 isoforms to determine the ratio between the 2 isoforms at each sample, and then to divide the normalized ADAR1 value of each sample to the 2 isoforms? If not- why, and what can I do to compare those similar isoforms?

Thanks,
Lea

[dpryan]

I wouldn't recommend DESeq (or even DESeq2) for looking at isoforms. Salmon and Kallisto can both be used with Sleuth and are more appropriate for your needs.

You can always use all three group, the individual comparisons then are just contrasts. Loading all three groups at once will likely produce slightly better results, in fact.

[Lea Shallev]

I didn't understand:
What do you mean by saying: " the individual comparisons then are just contrasts"? When looking at the gene level: is it possible to normalize the 3 groups together and then to compare pairs of groups?
Why do you say that loading the 3 groups at once produce slightly better results?

[dpryan]

Iis it possible to normalize the 3 groups together and then to compare pairs of groups?

Yes, this is called a "contrast".

Why do you say that loading the 3 groups at once produce slightly better results?

Having all the groups in allows for slightly better variance estimation (at least this is the case for most tools that you're likely ever to use).