Cuffdiff-confidence interval

[found]

Hi, everybody, I have a problem about cuffdiff. I used cuffdiff to test differential expression. However, I checked the "genes.fpkm_tracking". All the FPKM estimate have q1_conf_lo=0, and q1_conf_hi a big number, which means the 95% confidence interval are very large. It seems not reasonable. I am not sure if the test statistic is calculated by this intervals. If it is, then the number of the deferentially expressed genes would be smaller.

I also try to use cufflinks to estimate the expression level. and "genes.fpkm_tracking" produced by cufflinks have reasonable intervals. I remember cuffdiff used cufflinks quantification engine. So they should have similar result, right? Anyone have any ideas for this problems?

Thanks a lot

[IBseq]

did you get any answer?

ib

[IBseq]

did you find a solution? I have the same problem...

ib

[found]

No, I did not get reply.
However, I think the variance difference is just caused by numerical reasons. Since I do not have experimental replicate, using cufflinks to estimate the expression only consider the variance from the model, such as the variation resulting from the assignment a read to which isoform or possion distribution. this variance do not have any biological meaning. The true variance should include both the model variance and the biological replicate variance.

In cuffdiff, which compare expression difference from two samples/conditions. I believe it estimate the variance from two different samples/conditions. but in this case, the hypothesis that most genes have similar expression level is not necessarily hold. So I think that is the reason why the variance is larger in cuffdiff result.

What's your opinion?