Cuffdiff output shows log2(fold change) as infinite

[nr23]

Hi, I've compared two conditions in cuffdiff (2 reps each). On examining the output, I see in the gene_exp.diff file that some of the XLOC features show 'NOTEST' under the status column, and '-inf' under the log2(fold change) column.

I've never seen this before, and it's worrying me as it produces weird plots in cummeRbund...

Any help would be great!

Best,

N

[mattia]

log(zero)=-inf

I think you have 0 as a mean value (between replicates) in first condition (eg Treatment) and a number as mean value (between replicates) in the second condition(eg Control), so:
log2(FoldChange)=log2(meanTreatment/meanControl)=log2(0/number)=log2(0)=-inf

For "NOTEST", see the manual here in "Cuffdiff output" section.

[nr23]

So this isn't actually a bad thing, so to speak? If that's the case, I would have expected to see this sort of result quite frequently, where one condition has 0 reads mapping to a given gene and the other does have reads mapping.

I'm still worried/confused though, as when I run my results through cummerbund, and look for sigGenes, I find no genes with p<0.01, whereas looking through the gene_exp.diff I find many...

Do you know why this might be?

N

[sdriscoll]

When you're looking at the gene_exp.diff are you checking the q-value column or the p-value column? I think cummerbund uses the q-value column, which is correct.

Also +/- inf fold changes are totally normal. I think cuffdiff used to report either a very large number or a very small number in the past instead of 'inf'. So if you're used to using this program in the past then maybe you were used to those results?