Micro array: distribution of p-values

Hello all,

I'm having a (friendly) discussion with my PhD-promoter about the distribution of p-values:

Suppose you have the results of a micro array with expressions of n genes (eg. 10000) for t1 cases and t2 controls. One wants to compare gene expression between t1 and t2 so suppose one would do a t-test for every gene. This would result in n p-values.

Now the question is: what is the distribution of these p-values under the assumption that there is NO difference in gene expression. This would be possible by mixing up (permutating) the t1 cases and t2 controls and performing new t-tests (between t1' and t2').

Now I'm saying that these p-values are showing an uniform distribution because each p-value will occur by chance with a same probability. My promoter on the other hand is stating that, due to the permutations, the lower p-values will occur at a lower probability, thus not yielding an uniform distribution.

I'm almost sure I'm right ;-) but am I?

Thanks!
Tim
.

Relevant Pages

  • really easy integral, is actually really hard
    ... I am trying to model a distribution and calculate the expected ... after looking around for awhile at integral tables, ... Tim ... Prev by Date: ...
    (sci.math)
  • Re: really easy integral, is actually really hard
    ... On Tue, 9 Aug 2005, Tim wrote: ... I am trying to model a distribution and calculate the expected ... Classical Gamma function - Beta function exercise. ... Prev by Date: ...
    (sci.math)
  • Re: issues with statistical test suite from http://csrc.nist.gov/rng/
    ... > ses is the standard error of skewness. ... >>> this conforms to the expected distribution, ... with a proper test for the goodness of fit you are able to ... That final KS p-values seems really useless calculated that way, ...
    (sci.crypt)
  • Re: FFT test with few kbits
    ... > average to N/2 while the rest should average to zero. ... or -0.5 for the real components. ... > to test that they have the same gaussian distribution as the real ... p-values, and the same happens for the imaginary part. ...
    (sci.crypt)
  • Re: Interpreting the test result on a RNG
    ... of a random number generator. ... distribution as an approximation to the binomial distribution. ... CUMis the probability to see more than F failures. ... test of the hypothesis that the p-values are uniformly ...
    (sci.crypt)