[BioC] Factorial design with LIMMA

Gordon Smyth smyth at wehi.edu.au
Thu Apr 1 06:51:15 CEST 2004


At 02:23 PM 1/04/2004, cmprobst wrote:
>Hi,
>
>After using SAM for a long time, I have started an "struggle" with the 
>exceptional LIMMA package.
>
>After working with the example datasets and looking at the mailing list, I 
>have begun to analyse our own data.
>
>In the simpler experiments, I have not found any trouble and the package 
>did very well.
>
>But now I am trying to analyse a 2x2x2 factorial design, and I think I 
>have run into problems, with my biologist background.
>
>We are using Affymetrix Genechip, studying an infection process in two 
>different cell types and in two time points. There is two replicates for 
>each point.
>
>The phenoData slot is:
>
> > pData(resRMA)
>       Strain Infected Time
>Hyb01      A        1   6h
>Hyb02      A        1   6h
>Hyb03      A        1  24h
>Hyb04      A        1  24h
>Hyb05      A        0   6h
>Hyb06      A        0   6h
>Hyb07      A        0  24h
>Hyb08      A        0  24h
>Hyb09      B        1   6h
>Hyb10      B        1   6h
>Hyb11      B        1  24h
>Hyb12      B        1  24h
>Hyb13      B        0   6h
>Hyb14      B        0   6h
>Hyb15      B        0  24h
>Hyb16      B        0  24h
>
>I tried to create the following design matrix:
>
> > design<-model.matrix(~Strain*Infected*Time, data=pData(resRMA))
> > design
>       (Intercept) StrainB Infected1 Time6h StrainB:Infected1 
> StrainB:Time6h Infected1:Time6h StrainB:Infected1:Time6h
>Hyb01           1       0         1      1                 0 
>0                1                        0
>Hyb02           1       0         1      1                 0 
>0                1                        0
>Hyb03           1       0         1      0                 0 
>0                0                        0
>Hyb04           1       0         1      0                 0 
>0                0                        0
>Hyb05           1       0         0      1                 0 
>0                0                        0
>Hyb06           1       0         0      1                 0 
>0                0                        0
>Hyb07           1       0         0      0                 0 
>0                0                        0
>Hyb08           1       0         0      0                 0 
>0                0                        0
>Hyb09           1       1         1      1                 1 
>1                1                        1
>Hyb10           1       1         1      1                 1 
>1                1                        1
>Hyb11           1       1         1      0                 1 
>0                0                        0
>Hyb12           1       1         1      0                 1 
>0                0                        0
>Hyb13           1       1         0      1                 0 
>1                0                        0
>Hyb14           1       1         0      1                 0 
>1                0                        0
>Hyb15           1       1         0      0                 0 
>0                0                        0
>Hyb16           1       1         0      0                 0 
>0                0                        0
>attr(,"assign")
>[1] 0 1 2 3 4 5 6 7
>attr(,"contrasts")
>attr(,"contrasts")$Strain
>[1] "contr.treatment"
>attr(,"contrasts")$Infected
>[1] "contr.treatment"
>attr(,"contrasts")$Time
>[1] "contr.treatment"
>
>
>Whick looked very logical for me, but very complicated (well, I was 
>expecting something complex, anyway).
>
>So, before going into contrast analysis that could be meaningless, I 
>decide to ask for some advice from Bioconductor´s list.
>
>First, is this model correct?

Assuming that your strains A and B are different cell types, rather than 
biological replicates of the same cell line, then this looks a correct model.

>Second, I am interested in several aspects (contrasts), which I can 
>address if asked:
>
>For instance, differences between cell types without infection, and 
>differences between cell types with infection (time excluded or included).
>
>Which contrasts can answer these questions?

Ah, this is the big question. I hope someone other than me will jump in 
here, because finding interpreting contrasts from factorial models is not 
specific to limma.

>  How many constrasts I can analyse? All of them?

Yes.

>  Is there sufficient degree of freedom?

Yes.

Gordon

>Thanks in advance for your assistance.
>
>Christian Probst
>Bioinformatics - IBMP



More information about the Bioconductor mailing list