Identifying Genes That Express Different Isoforms In Cancer Vs Normal Rna-Seq Data

Question

[Sahel]

Hi There,

I am trying to find genes that express different isoforms in cancer vs normal RNA-seq data. The first thing that came to my mind was to do a differential expression analysis at exon level (I used edgeR for this)... then if I find a gene that express less number of exon compared to the annotation data for that gene, it means the gene is expressing a different isoform in cancer.

But this is clearly not the right way to do this, and I could not find any other way for this kind of analysis. I was wondering if someone has done this kind of analysis before can share his/her experience with me.

I really appreciate your help, Thanks, Sahel

Answer 1

[Ahmetz]

Sahel, you should check you DEXseq package. It is made for detecting differential exon usage and has nice visualizations.

[Sahel]

This is great, thank you so much Ahmet :)

[beroe1]

This is an old thread, but for what it is worth the DEXseq link is now here.

Answer 2

[JC]

Maybe the Tuxedo-toolkit can help you: Bowtie + TopHat (with fusion detection) + Cufflinks (with novel isoforms detection). I will run the same analysis in both samples, then with cuffdiff select the interesting isoforms.

[Sahel]

Hi JC, I actually took your advice and started to use Tophat/Cufflink a few weeks ago. I have 15 tumor and 5 matched normals (human RNA-seq). Everything works great except Cuffdiff. Right now I am running Cuffdiff on the merge.gtf file of all 20 samples, but it keeps running out of memory and looks to take weeks to run! (I am using 12 cpus with 50GB RAM). I was wondering if you experience the same thing or I am doing something wrong? I have another 100 sample coming in for this project, however it seems it is impossible to use tophat/cufflink with 100 samples! Thanks very much, :-)

[JC]

looks like a problem in the cufflinks installation, I didn't see that error in my analysis previously. You already have the assembled GTF, so you can use edgeR or DEXseq with the reads counts if you cannot fix the cuffdiff problem.

Answer 3

[Kristoffer Vitting-Seerup]

An alternative IsoformSwitchAnalyzeR ( http://bioconductor.org/packages/IsoformSwitchAnalyzeR/ ) which enables statistical identification of isoform switches with predicted functional consequences where the consequences can be chosen from a long list but includes gain/loss of protein domains, signal peptides changes in NMD sensitivity etc. IsoformSwitchAnalyzeR directly parses RSEM output (also support cufflinks/cuffdiff, Kallisto andSalmon). IsoformSwitchAnalyzeR

Hope this helps

Kristoffer