cuffdiff (v 2.1.1) p-values

[nsl]

I have rnaseq data which I analyzed using the latest versions of tophat-cufflinks-cuffdiff ( that includes the new dispersion methods)

The authors mention that, "Because the test is now based on explicit sampling from the beta negative binomial, users will not see values less than 10^-5 by default" .

I understand this is the lowest possible p one will see, which then in turn outputs a uniform adj -p (0.00801514) for 45 of my genes that had equal or smaller than 10^ -5 p-val.

Is it legitimate to see the same p val for 100's of genes when it is NOT equal or less than 10^-5?

Thanks

[janos]

I was wondering about the same issue.
Looks as if the genes are sorted to discrete p/q bins, and every 4-5 or so genes (in my case) appear to have the same p/q values.

Janos

[nicrob]

As of version 2.1, cuffdiff uses a sampling method to calculate the p-values. As I understand it, the new method draws about 10000 simulations from the null hypothesis of no significant change in expression, and defines the p-value as the proportion of those 10000 simulations that were more extreme than what was observed. As a consequence, the p-values you get out are slightly 'discretised', and many sites may have the same p-value (the p-values can only be fractions out of 10000.)

Hope that answers your question.

[janos]

Yes, that answered my question, thank you nicrob!

[nsl]

Thanks you both for the insight