Different FPKM values of cufflinks and cuffdiff

[IBseq]

hi robby, not sure if you sorted out but I write it anyway...

"Maybe I do something wrong. I did the following:
1. cufflinks -G ucsc-reference.gtf sample_n.bam (for all samples)
2. cuffdiff ucsc-reference.gtf sample_1.bam sample_2.bam [....] sample_n.bam
3. compare genes.expr (cufflinks) with genes.fpkm_tracking (cuffdiff)
"

you run cufflinks correctly, but then you need to run cuffcompare and then use the output of cuffcompare to run cuffdiff to which you will provide as well the sam/.bam files from tophat...

hope it helps,
ib

[Robby]

Thanks for the fast reply, mrfox.

Maybe I do something wrong. I did the following:
1. cufflinks -G ucsc-reference.gtf sample_n.bam (for all samples)
2. cuffdiff ucsc-reference.gtf sample_1.bam sample_2.bam [....] sample_n.bam
3. compare genes.expr (cufflinks) with genes.fpkm_tracking (cuffdiff)

@mrfox: Which values did you compare with each other? Did you perform the same steps described above?

[mrfox]

Hi all, I used v9.3 now and cufflinks and cuffdiff generate consistent values.

[Robby]

Hi all, I use version 9.3, but I still have different FPKM values returned by cufflinks and cuffdiff. Did anybody observe similar problems? Has anybody tested, which results are more trustable?

Please let me know, if somebody has any updates or recommendations. Thank you all.

[fatrabbit]

hi, I also faced the same problem by using version 9.1. I'm trying to use version 9.3 (the latest one), and to see if it still has the problem.

In their website, they said "This release fixes several issues that affect abundance estimation and differential expression accuracy, and is strongly recommended for all users."

If you have any updates, please let me know. Thanks.

[mrfox]

Hi all, sorry for repeatedly posting the same question. But I am still confused about the different FPKM values returned by cufflinks and cuffdiff. Did I make mistakes in one of the steps? Below is how I use the package:

1)run cufflinks on the sam files obtained by Tophat
2)run cuffcompare on the gtf files in last step, I used UCSC known genes as the reference. And I got the stdout.combined.gtf and the tracking files as expected.
3)using the stdout.combined.gtf obtained in 2) as reference, I run cuffdiff on the sam files of the samples and obtained the *.diff and *.tracking files.

However, I found that for many isoforms (from isoforms.tracking), their FPKM values from the samples usually do not match those values by cufflinks from step 1). The version I used is 0.9.2.

Could anybody help to explain this? And, what is the reason to re-assign the reads to transcript and re-estimate abundance in cuffdiff?

I desperately look forward to your answers. Thank you all.