DESeq2 for isoform level

[Golsheed]

Hello,

(1) I was wondering whether DESeq2 is appropriate to test for differentially expressed isoforms? If we have raw counts (on isoform level), can we treat them like gene expression levels and run DESeq2?

(2) Another question and a more specific one is: my isoforms counts are actually the estimated outputs of eXpress (eff_counts). Here's the explanation:

"The estimated number of fragments generated from this target in the sequencing experiment, adjusted for fragment and length biases. In other words, his is the expected number of reads from the experiment if these biases did not exist. This is the value recommended for input to count-biased differential expression tools."

I'm assuming that this can't be used as an input to DESeq2 since it's corrected for some biases, is that correct? If so, then is there another package that can be used for this?

(3) In fact, eXpress has another output (est_counts), explained as follows:
"The estimated number of fragments generated from this target in the sequencing experiment."
I'm guessing this is the actual estimated counts, without any bias correction ... would this be appropriate for isoform-level DE analysis with DESeq2?

Thanks in advance for any sort of comment and suggestion!

Golsheed

[Michael Love]

hi Golsheed,

The question about isoform counts has been asked and answered many times on Seqanswers and elsewhere, here is one of Simon's answers: https://support.bioconductor.org/p/43395/#43400 . DESeq2 is only designed for gene-level analysis.

[Golsheed]

DESeq isoform level

Thanks a lot for helping, Michael.
(1) I guess the reason I got confused was that after reading Simon's answer, I also read this one:

https://stat.ethz.ch/pipermail/bioconductor/2014-May/059479.html

What are your thoughts on this? Since the corrected counts that I have at hand are from the output of the eXpress software, would this change your answer?

(2) If not, would it be wrong to simply use a linear model to detect differential isoform expression, with the corrected counts I have at hand from eXpress? or does it have to be done with some of the available packages from scratch?

I appreciate your help very much,
Golsheed

[Michael Love]

In my answer there, I am saying that the uncertainty of the isoform-level estimates from assigning shared reads is not passed through in the value, so we don't recommend this use for DESeq2. The Huber group does have DEXSeq for detecting differential splicing, using raw counts. There are plenty of methods and I'm not the one to say which you should use, other than to advise against isoform-level analysis with a gene-level tool.

[Golsheed]

Makes sense. Thanks a lot for always being helpful.

Golsheed