[BioC] Limma - Design related problem: Unconnected design and time-treatment interaction estimation

[Serge Eifes]

Dear all,
I’ve got the following microarray experiment design-related problem 
when using Limma: In our lab we want to perform some microarray
hybridizations to identify genes whose expression is altered by a 
certain treatment over a time. We want to do cohybridizations of RNA
from treated vs untreated cells at corresponding time points to
spotted cDNA arrays. This time series should be performed in
triplicate, including three biological replicated and a dye-swap

Tx=treated, Ux=untreated cells at  time point x
* T2 – U2
* T6 – U6
* T12 – U12

My problem is the following: Is it statistically valid to
estimate  the interaction Time:Treatment using such an unconnected
design?

Using the following design matrix

      T2vsU2 T6vsU6 T12vsU12
 [1,]      1      0        0 
 [2,]      1      0        0  
 [3,]     -1      0        0   
 [4,]      0      1        0 
 [5,]      0      1        0        
 [6,]      0     -1        0 
 [7,]      0      0        1
 [8,]      0      0        1
 [9,]      0      0       -1

and contrast matrix 

          Contrasts
Levels     Interaction_6_and_2_hours Interaction_12_and_6_hours 
  T2vsU2                      -1                      0
  T6vsU6                       1                     -1
  T12vsU12                     0                      1 

the script can be executed without error message using R (vers. 2.4.0)
and Limma (2.9.8), while the same kind of analysis in LimmaGUI (1.10.0)
produces the following error message :

“For this experiment, all possible RNA comparisons can be obtained from
one parameterization.
This is either because there are only 2 RNA types, or there is an
unconnected experimental design and there only 2 RNA types within each
sub-experiment.”

A former post from 2006-10-27 by Richard Friedman was already related
to the same kind of error message in LimmaGUI. What I wanted to know
is wether this message is specific to LimmaGUI and its implementation
as compared to Limma or does it point out that this design is not
appropriate for this kind of analysis as in  Limma no error message is
displayed.

I would appreciate any comment or suggestion.

Best regards
Serge


Serge Eifes
Laboratoire de Biologie Moleculaire et Cellulaire du Cancer (LBMCC)
Hopital Kirchberg 
9,rue Edward steichen 
L-2540 LUXEMBOURG
Phone:+ 352 2468-4046	 Fax : + 352 2468-4060