differentiated expression analysis on isoforms/transcripts level

[arrchi]

Hi there,

I have a RNA-seq data (3 cases and 4 controls). I am looking for ways to identify differentiated isoforms/transcripts in these two conditions. Any suggestions besides t-test?

Thanks.

[gringer]

DESeq and EdgeR have been frequently recommended in this forum

[arrchi]

As far as I understand, DESeq and EdgeR identify differentiated genes, but not transcripts/isoforms. I wish they have extended their software's ability to isoform level.

[peromhc]

Cufflinks is probably your best bet for isoform level analyses.

[kopi-o]

ALEXA-SEQ could also be worth looking into.

And DEXSeq looks at differential exon usage.

[Jane M]

There are also IsoEM and, a bit different, DEXSeq

[JSeed]

"As far as I understand, DESeq and EdgeR identify differentiated genes, but not transcripts/isoforms."

This doesn't make sense to me. These programs analyze differential expression, whether that occurs at the gene level or at the transcript isoform level shouldn't matter. The mapping program that you use to identify splice junctions will help sort out the various isoforms and some mappers are better than others. Commonly used programs include ERANGE and SpliceMap among others. Bowtie/Tophat is not my favorite because it just doesn't perform quite as well as some other aligners and you end up locked into Cufflinks, which is not necessarily the best option. Whatever you do, don't just perform a t-test! Some useful recent reading includes Risso, et al., doi:10.1186/1471-2105-12-480, or Tarazona, et al., doi:10.1101/gr.124321.111 or several others on this topic over the last 6 months.

[arrchi]

Thanks for all responses.

That's right. Cufflink-cuffcompare-cuffdiff gives results of paired samples, but not between two conditions.

I am reading the paper and the manual of "edgeR", you might right. the authors said their method might apply to transcript level. Then I wonder how to get raw count from RNA-seq fastq files. Also the cpm function in edgeR is missing. Anybody know what it is?

Thanks again.

[arrchi]

I found cpm in their source package.

[kopi-o]

The problem is that it's hard to assign integer read counts to transcripts, and edgeR / DESeq work with integers. All the isoform/transcript quantification models that I know of - most are based on EM, like Cufflinks, although there are also linear equation system models etc. - probabilistically assign reads to multiple isoforms with different non-integer weights. One way to circumvent this is to only consider reads that map to a single isoform only - it may be that e g ALEXA-SEQ and NEUMA work in this way. That would achieve integer counts without double-counting reads.

[arrchi]

Is it a reason that we can not fit a GLM model with FPKM value directly since FPKM is a probability?

[arrchi]

found some helpful information here in case one is wondering this type of question like me:http://seqanswers.com/forums/showthread.php?t=4349

[DineshCyanam]

That's right. Cufflink-cuffcompare-cuffdiff gives results of paired samples, but not between two conditions.

That is wrong. cuffdiff can be used to compute the differential expression between two conditions. For two conditions, you need to give the sample names as comma separated values and differentiate the conditions with a space.

cuffdiff --labels treated,untreated cuffcompare_combined.gtf treated_1.sam,treated_2.sam,treated_3.sam untreated_1.sam,untreated_2.sam

The cuffcompare_combined.gtf is obtained by running cuffcompare with the above 5 samples together.