How do you calculate standard deviation on replicate RNA-seq data?

[kwatts59]

To all,
I extracted RNA from 3 untreated samples and had RNA-seq analysis performed.
A couple months later, I extracted RNA from 3 treated samples and had RNA-seq analysis performed.

I ran tophat and cuffdiff to generate the differential expression (DE) data between the 3 control samples and the 3 treated samples.

How do I calculate the standard deviation for the DE for the genes of interest on this data? Cuffdiff only provides a p-value and q-value.

Thanks in advance.

[mastal]

How do How do you calculate standard deviation on replicate RNA-seq data?

Cuffdiff also gives the value of the test statistic used to calculate the p-values.

You do realize that since all your untreated samples were extracted and run at one time, and all your treated samples were extracted and run at a different time, this will make it difficult to establish whether any differences between treated and untreated are due to the treatment, or to the fact that they were processed and run at different times.

[eneella]

Kwatts,

I don't know if you ever resolved this, but it never hurts to post a reply, in case someone else has the same issue and can't find the answer.
I haven't done it myself, but I was told to run Cuffdiff twice - once with comma delimiters to denote replicates, and once without. Then you can take the data from the individually analyzed samples and calculate the standard deviation.

I hope it works for both of us!

[dpryan]

Kwatts,

I don't know if you ever resolved this, but it never hurts to post a reply, in case someone else has the same issue and can't find the answer.
I haven't done it myself, but I was told to run Cuffdiff twice - once with comma delimiters to denote replicates, and once without. Then you can take the data from the individually analyzed samples and calculate the standard deviation.

I hope it works for both of us!

Out of curiosity, why do you want to do that?

[sindrle]

Im thinking the same thing.... I thought in RNAseq, since its modelled after the negative binomial model, the variation is actually the dispersion = squared CoV and CoV = mean/SD.

But this is the common dispersion, in edgeR and DEseq they use tag wise, but you should not use tag wise dispersion per gene (setting prior.df = 0), since its unreliable. Conclusion is, you have no measurement of the per gene variation.

So even if you calculate SD from Cuffdiff 2.1.1 on a single gene, this value will not mean anything, because it has nothing to do with the testing that has been done.


My supervisor still demands me to calculate SD...