Cuffdiff Strange Output

Question

[jenny.ditommaso]

Hi, I hope that someone can be so gentle to help me. I'm analizing some data from RNA-Seq with TopHat and Cufflinks and I focus my attention on differently spliced genes. In particular, now I'm working on "maybe false positive", for example:

NM_001130171 Zfp207 Zfp207 chr11:80196780-80219375 day0 day3 OK 0.0000000 36.6325000 1.79769e+308 1.79769e+308 2.67658e-19 1.72370e-18 yes

this row comes from isoform_exp.diff file and the log2fold change is totally crazy because one sample is zero. So, I control the expression with expressionBarplot from cummeRbund, and the fpkm value of sample one was not zero! This function takes the information from another file, gene_exp.diff:

Zfp207 chr11:80196780-80219375 day0 day3 OK 26.4421 97.6658 1.88502 -15.6715 0 0 yes

The two files should be have the same results, but they are different and a p-value=0 is really strange.

Have someone some information about this? There is some normalization not explained in Cufflinks web site? What is the final true fpkm value?

Thanks :)

Answer 1

[Fabio Marroni]

Nothing strange, actually. The log2fold change is a very large number because log N/0 = Inf. They just decided to approx Inf with a very large number. If you use gene_exp.diff you can obtain different expression values than isoform_exp.diff (even for the same gene) if cufflinks finds some evidence of alternative splicing, so no surprise. On the cufflinks website they explain their normalization method. Good luck!

Answer 2

[jenny.ditommaso]

Thank you for the answer.