Re: permutation p value vs. normal p values

On 16 Aug 2005 11:09:50 -0700, arrayprofile@xxxxxxxxx wrote:

> Hi, I am performing Cox proportional hazards regression on a microarray
> dataset with 15000 genes. The p values generated from the Cox
> regression based on normal distribution showed only 2 genes have a p
> value less than 0.05. However, I did a permutation on the dataset to
> obtained permutated p values, and it turned out about 750 genes had a
> permutated p value less than 0.05. With that big difference in the
> number of significant genes, which one should I trusted? and what's
> reason why such a big difference exists? My dataset is not large in
> sample size (17 samples), might this be the reason?


Having a data set with no-power for the test in
hand will result in having no 'significant' results.
For instance, you can't get a non-random selection
of head/tails if you looking at sets of 3 flips.

Off hand, I don't have any idea what you might be
using a Cox proportional hazards model for, with
microarray data. Where do you get a time-related
event, anyway? Were you creating dichotomies from
each array item, and looking at some survivorship time?

Having 750 p-values at 5% or less, out of 15000 tests,
is exactly the number expected by chance. Is this, indeed,
the actual number, and not a mis-reading?
-- "permutions" takes the extreme 5% of results as a marker,
or cutoff. If there was exactly no-effect, the permutation cutoff
would also cutoff 5% of the sample, or 750 plus or minus 55
or so. I'm wondering if this is coincidence, or if you might be
looking at the wrong numbers, somewhere.

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.

Relevant Pages

  • Re: Path from a walk in a graph.
    ... The basic principle of the genetic algorithm is to take a non optimal ... variables which we can loosely describe as being genes. ... graphs these will be permutation subgroups. ... completeness is proved by showing that no permutation or set of ...
    (sci.math)
  • permutation p value vs. normal p values
    ... I am performing Cox proportional hazards regression on a microarray ... dataset with 15000 genes. ... I did a permutation on the dataset to ... might this be the reason? ...
    (sci.stat.consult)