[BioC] decideTests with nestedF

[Gordon Smyth]

>[BioC] decideTests with nestedF
>Lei Jiang lj22 at u.washington.edu
>Wed Jul 27 22:04:21 CEST 2005
>
>Dear all,
>
>I am comparing two time points (6 time points) data under different 
>conditions.
>I want to look for genes that show different pattern under different
>conditions.
>
>I constructed contrasts as in the user's guide (diff=cond1-cond2 at all pairs
>of time points) and generated MArrayLM object okay.
>
>My questions come with decideTests with "nestedF".
>1. how is the classification done exactly??

The method is described in ?classifyTestsF, 3rd paragraph of Details.

The "nestedF" method is my research and has not yet appeared in a published 
paper that I can refer you to. If you're not comfortable with this, use one 
of the other methods instead, e.g., use the default method.

>  I found a much smaller set of genes
>using decideTests(nestedF, p=0.01) versus p.adjust(fit$F.p.value)<0.01

This is not possible, so I think you might not have used the methods 
correctly. Do you understand that F.p.value<0.01 implies only that *one* of 
the contrasts will be significant, not that all will be? Can I suggest that 
you explicity specify the adjustment method argument whenever you use 
decideTests() and p.adjust() to make sure that you are using the same 
multiple testing adjustment method for each. Otherwise you're comparing 
apples with oranges.

>2. The TestResults matrix doesnt really make sense. for example, places where
>the matrix shows 1 as upregulation, dont necessarily have positive M value or
>doesnt show a p.value < 0.01?? What do I miss here?? really confused...

This is not possible, so either you've made a mistake or I have. If you can 
produce a reproducible example, I will check it. Note that F.p.value<0.01 
is required but p.value<0.01 is not.

>3. p.value and adjust.method parameter in the decideTests are applied to F 
>stat
>or to the individual p value at each gene at each contrasts.

This depends on the 'method'. If 'method="nestedF"', then the adjust.method 
is applied to F.p.value. "nestedF" has its own method of handling the 
individual contrasts.

Gordon

>I hope I made the questions clear. Need help desperately, since there's not
>much document on this. Other suggestions for this type of analysis are also
>welcome.
>
>Thank you.
>
>Lei Jiang
>
>
>Department of Chemsitry
>University of Washington
>Box 351700
>Seattle, WA 98195
>Phone: 206-616-6882
>Fax: 206-685-8665