[BioC] globaltest question

[Goeman, J.J. (MSTAT)]

 Dear Mike,

The permutation version of globaltest is safe but conservative for small
sample size. It can always be used, even in small groups, but it is not
so useful if you want to test many pathways because a Bonferroni or FDR
correction may leave you with no significant pathways at all due to the
conservatism of the permutation test.

In that case you may therefore want to use the asymptotic version. This
is like using the t-test for small samples when you are not completely
sure that the data are normally distributed, so some care should be
taken when interpreting the results. But for mining pathways for strong
association with your phenotype this works quite well. Except in unusual
situations, the most asymptotically significant pathways will also have
the smallest possible permutation p-value.

For multiple testing correction see the gt.multtest function.

Jelle



-----Oorspronkelijk bericht-----
Van: mike Ad. [mailto:mikeaddr at hotmail.com] 
Verzonden: dinsdag 12 september 2006 20:57
Aan: Oosting, J. (PATH); bioconductor at stat.math.ethz.ch
Onderwerp: Re: [BioC] globaltest question

Hi,
Thanks for the reply!
I have two following questions:
1. It is ok to use globaltest for small groups? (totally 10 arrays for 2

groups in my case.)
2. Different method ("auto", "asymptotic"...) in the globaltest gives 
different p-values, how should one set threshold to pick out the
significant 
pathways?
Thanks,
/Mike




>From: "Oosting, J. (PATH)" <J.Oosting at lumc.nl>
>To: "mike Ad." <mikeaddr at hotmail.com>,<bioconductor at stat.math.ethz.ch>
>Subject: RE: [BioC] globaltest question
>Date: Tue, 12 Sep 2006 16:42:25 +0200
>
>On small groups the globaltest automatically uses permutation tests.
You
>can see in your result that in this case there are 210 permutations,
and
>all p-values will therefore be multiples of 1/210, with a minimum of
>1/210(=0.0047619). You can force it to use a more gliding scale by
using
>the argument method="asymptotic".
>
>Jan
>
>-----Original Message-----
>From: bioconductor-bounces at stat.math.ethz.ch
>[mailto:bioconductor-bounces at stat.math.ethz.ch] On Behalf Of mike Ad.
>Sent: dinsdag 12 september 2006 16:23
>To: bioconductor at stat.math.ethz.ch
>Subject: [BioC] globaltest question
>
>Dear list,
>
>I am new to use the "globaltest" packages (version "4.2.0"). I have 10
>mouse arrays from two groups (control and treated). I tested them
>against all the kegg pathways. The result looks stage to me because
>among the 171 pathways tested, most of them have the identical p-value.
>And that p-value is the smallest.
>The code I used is listed, could someone help to tell me where went
>wrong with my code?
>
>Thanks!
>
>/Mike
>
>kegg<-as.list(mouse4302PATH2PROBE)
>
>gtkegg<-globaltest(affy_expression, diagno, kegg) ##where the first
>argument "affy_expression" is the affy expression data set I got by
>using function "exprs()", each row is one affy probe and each column is
>from one array.
>## the second argument "diagno" is a vector containing 10 group names
>("treated" or "control") for the 10 arrays and they are in the
>corresponding order to the 10 columns in the expression data.
>
>gtkegg<-sort(gtkegg)
>
>#Just list the top 5 of the result, the P-value are identical, what's
>wrong?
>gtkegg[1:5]
>Global Test result:
>Data: 10 samples with 45101 genes; 5 pathways tested
>Model: logistic
>Method: All 210 permutations
>
>       Genes Tested Statistic Q Expected Q sd of Q   P-value
>00623    12     12      37.552     9.1318  9.2135 0.0047619
>00440    47     47      13.010     3.3143  1.7585 0.0047619
>00624    43     43      57.812     9.1819  8.2350 0.0047619
>00625    19     19      71.404    12.6820 10.4620 0.0047619
>00626    28     28      15.648     3.5587  2.2039 0.0047619
>
>_________________________________________________________________
>Check the weather nationwide with MSN Search: Try it now!
>

_________________________________________________________________
Windows Live Spaces is here! It's easy to create your own personal Web
site. 
  http://spaces.live.com/signup.aspx