Differences in FPKM values for same sample in multiple comparisons

Question

[jammydodger123456]

I have 7 samples (call them A,B,C,D,E,F,G). I want to compare all of the samples against sample A.

When I run cuffdiff for A against B, A against C .... and A against G, the FPKM for gene X for sample A in the AvB comparison will be different to the FPKM value for gene X for sample A in the AvC comparison etc.

Has anyone else noticed this in their own analyses? Should it be concerning?

[Kristoffer Vitting-Seerup]

It happens because Cufflinks/Cuffdiff uses a statistical model where it considers all the data you supply. Since you provide differen input data the results are different. Use the contrast function as suggested by @Satya below.

[Satyajeet Khare]

Does this happen if you provide all comparisons together in the form of a contrast file with single cuffdiff command or when each comparison is tried with a separate cuffdiff command?

[jammydodger123456]

It happens when I do separate commands. I'm not sure how to do those kind of comparisons in one command. Would you be able to show me the general command for that kind of comparison, or point me in the direction of any help? Thanks

Answer 1

[Satyajeet Khare]

This is a typical cuffdiff command with contrast file.

cuffdiff --use-sample-sheet -C contrast.txt -p 10 -o cuffdiff_out Gene.gtf sample_sheet.txt

This does not change FPKMs in my experience.

[jammydodger123456]

In your opinion, where there are multiple replicates for each sample, would you recommend running them all in one Tophat command prior to this, so that there is a single .bam file for each sample? If not would you advise on how the sample sheet and contrast.txt should be presented? Thanks for your continuing help

[Satyajeet Khare]

If you have multiple technical replicates, you can combine them into one .bam file. If you have biological replicates, they should better be kept separate. They can be grouped by providing a group name as a column in sample sheet. The contrast file will then compare groups one-by-one.

Answer 2

[jammydodger123456]

Thank you Kristoffer and Satya