Cuffdiff and zero FPKM values give enormous log ratio

[altodor]

Hi all,
I'm new to RNA seq and I'm working with TopHat - Cufflinks - Cuffcompare - Cuffdiff pipeline.
I use Cufflinks without annotation and then Cuffcompare for two samples with annotation. I put the results of Cuffcompare (transcripts.combined.gtf) to Cuffdiff as input. Is this the right way?

I have results from cuffdiff for differential expression testing. As a result for DE testing it provides the following measure: ln(FPKMx/FPKVy) for ratio for the gene between two samples X and Y. If for gene A in sample X FPKMx = 2 and for the sample Y FPKMy=0, then the ratio is like 1,79 E306, so enormously big
How do you deal with it?

Any ideas are very welcome!

[dahlo]

Hello

Regarding the huge ratios, i think they are hard to avoid. There is no other way to handle the scenarios where you have a gene that is switched on/off, that i know of.

If you were to plot your ratio values, you would likely see most of them around 0 - a couple of thousand perhaps, and some scattered outliers with ridiculous high values. You can then set a suitable cutoff and say that all genes with a ratio higher than X (say 1e10 or 1e20) are genes that have been turned off or on. It is pointless to sort them with respect to their ratio size, since they all should have Inf.

Computers don't like to divide by 0, so a small number (like 1e-500 or so) has been added to the FPKMy in the division to keep the program from crashing, since many programming languages can't handle infinite numbers.

Cheers
Dahlo

[altodor]

Thank you!