RNA-seq determination of expression

[irholtman]

Hi all,

I would like to determine whether RNA-seq RPKM values are expressed in a particular cell type or not. Let's say sort of a one sided t-test, which of course is not feasible because there are not so many replicates (N=2 or 3 per condition), and get p-values to determine the significance of expression.

I used tophat, cufflinks and cuffdiff to align data and calculate the RPKM values per sample, however, since I have 2 or 3 values per cell type and would like to determine the expression values based upon all samples, I don't know how to calculate these.

Cuffdiff gives an 95 % confidence interval around around the FPKM values per cell type, but these confidence intervals are not multiple testing corrected, and do not contain p-values.

Does anybody have an idea how I can test which genes are expressed in a particular cell type with the asssociated multiple testing corrected p-values based upon RNA-seq data?

Many thanks!

Kind regards,
IR

[dpryan]

Plot the RPKM distribution and see if it has more than one peak. It's likely that you have a peak ~0 and then another one higher up, so you can fit it (probably with a gaussian) and use that as the distribution against which to do a single-tailed T-test. You can adjust the resulting p-values for multiple comparisons.

[irholtman]

Thanks for your response. I have plotted it but many values are zero. But the normal distribution for most samples only applies to the log RPKM values. There is not a normal distribution (or any other peak), just above the zero, with the exception of many genes beeing exactly zero, which is not distribution.

Are there any bioconductor packages or linux tools that I could use to calcute significance of expression?

Regards,
IR