[BioC] SAM for protemics - low q values

[Hendrik Nolte]

Hi all,

usually I analyze my MS data (around 3,000 proteins, 4 replicates (too
little?) using benjamini hochberg correction for multiple testing. (or
permutation based q value calculation however the comp.fdr function (DEDS
package) gives me different raw p value then i get using normal t.test
(same properties uneq var, unpaired) - they seem to be already corrected
(same value appears multiple times)) Anyway i was wondering whether i can
use the SAM package as it is? And furthermore, why the max q value is 0.35
(and not 1?). Shouldnt be there proteins/genes that would be called
significantly changed if we had applied a FDR of 1. Sorry if this question
is kinda silly ..

SAM Analysis for the Two-Class Unpaired Case Assuming Unequal Variances

 s0 = 0.0382  (The 10 % quantile of the s values.)

 Number of permutations: 70 (complete permutation)

 MEAN number of falsely called variables is computed.

   Delta    p0    False Called    FDR  cutlow  cutup   j2   j1
1    0.1 0.407 2433.900   2781 0.3563  -0.222  0.170 1354 1623
2    1.1 0.407   52.900    581 0.0371  -2.095  2.199  327 2796
3    2.2 0.407    3.800     96 0.0161  -3.866  3.881   50 3004
4    3.2 0.407    1.286     36 0.0145  -5.316  5.120   16 3030
5    4.2 0.407    0.514     15 0.0140  -6.751  6.412    7 3042
6    5.3 0.407    0.186      5 0.0151  -8.386 10.158    4 3049
7    6.3 0.407    0.157      4 0.0160  -8.985 10.158    3 3049
8    7.3 0.407    0.057      2 0.0116 -11.773    Inf    2 3050
9    8.4 0.407    0.057      2 0.0116 -11.773    Inf    2 3050
10   9.4 0.407    0.043      1 0.0174 -12.454    Inf    1 3050

For a FDR cutoff of o.05 i can either use the find delta function or just
pick 1.1 to be on the safe site.


BEsts