[BioC] Genechip experiment in limma

[Gordon Smyth]

At 02:48 AM 22/08/2003, Christian Probst wrote:
>Gordon,
>
>I succeed in loading the cDNA microarray data. As I need a spatial
>normalization, I will try to do it before the use of limma package.

Remember that print-tip-loess is in itself a simple form of spatial 
normalization (as well as intensity-based normalization). In my experience, 
print-tip-loess does most of what spatial normalization would have done. 
But, as you say, if you really do want to do formal 2D normalization you'll 
have to do it outside of limma.

>I am also running a Genechip experiment, with four time points and three
>treatments (A,B and Ctrl), with no replicates.

Am I correct in assuming that your first 4 chips are hybridised with Ctrl, 
the second four with A and the last four with B? Are the time points 
1,2,3,4,1,2,3,4,1,2,3,4?

>I would like to ask you about the design table in order to analyse this
>data.
>I want to find those genes that are differentially expressed in AxCtrl,
>BxCtrl, and AB x Ctrl.

Do you want genes which are DE between A and Ctrl and DE between B and 
Ctrl? Obviously this includes identifying genes which are both DE for A vs 
Ctrl and B vs Ctrl. I am not sure what you mean by "AB x Ctrl", perhaps you 
just mean different for both A vs Ctrl and B vs Ctrl.

>Is this design matrix correct?

No it's not. If my assumptions above are correct, and your time points were 
really replicates, which they're not, then you'd want either

 > design
Ctrl A B
1 0 0
1 0 0
1 0 0
1 0 0
0 1 0
0 1 0
0 1 0
0 1 0
0 0 1
0 0 1
0 0 1
0 0 1

with contrast matrix

 > cont.matrix
A-Ctrl  B-Ctrl
-1    -1
  1     0
  0     1

 > design
Ctrl A-Ctrl B-Ctrl
1 0 0
1 0 0
1 0 0
1 0 0
1 1 0
1 1 0
1 1 0
1 1 0
1 0 1
1 0 1
1 0 1
1 0 1

with contrast matrix

 > cont.matrix
A-Ctrl  B-Ctrl
0     0
1     0
0     1

I suggest using the first design matrix, in which case using limma 1.1.11, 
you would go

fit <- lmFit(eset, design)
fit <- contrasts.fit(fit, cont.matrix)
fit <- eBayes(fit)
clas <- classifyTests(fit)
vennDiagram(clas)

to classify each gene as differentially expressed or not for AvsCtrl, 
BvsCtrl or both.

> >design
>      A  B AB
>1    0  0  0
>2    0  0  0
>3    0  0  0
>4    0  0  0
>5    1  0  1
>6    1  0  1
>7    1  0  1
>8    1  0  1
>9    0  1  1
>10   0  1  1
>11   0  1  1
>12   0  1  1
>
>As I have a time-course experiment, and I am using the four time points
>as replicates, in order to analyze this data statistically, I wonder if
>there is a design table that uses the temporal information in a flexible
>way, to identify DEG in the beggining or ending of the infection
>proccess (I will mix my "replication" with my "temporal framework").

Yes, there is a design matrix which incorporates the temporal information, 
but it is difficult to do statistical analysis in the absence of any 
replication. (I do wish people would be prepared to invest in an extra two 
or three chips in order to be able to do statistical analysis.)

There are things you could do, for example using genes which are not 
differentially expressed to estimate variability for genes which are 
differentially expressed, but this is a research problem.

Regards
Gordon

>TIA
>
>Christian
>
>
>-----Mensagem original-----
>De: Gordon Smyth [mailto:smyth at wehi.edu.au]
>Enviada em: terça-feira, 19 de agosto de 2003 23:01
>Para: Christian M. Probst
>Assunto: Re: Limma error
>
>Christian,
>
>At 02:50 AM 20/08/2003, Christian M. Probst wrote:
> >Dear Mr. Smyth,
> >
> >I am trying to use the limma package, but in the first step of the data
> >loading, I run into the following error:
> >
> > > library(limma)
> > > files<-dir(pattern="*\\.spot")
> > > files
> >[1] "EP3_05_ET5_05_0808.spot" "EP3_28_ET5_28_0908.spot"
> >[3] "SP3_30_EP5_28_0908.spot" "SP3_30_ST5_30_0908.spot"
> > > RG<-read.maimages(files,source="spot")
> >Read EP3_05_ET5_05_0808.spot
> >Read EP3_28_ET5_28_0908.spot
> >Error in "[<-"(*tmp*, , i, value = NULL) :
> >         number of items to replace is not a multiple of replacement
>length
> > >
> >
> >This message appears consistently in other files, also. I havent found
>a
> >description in the Bioconductor discussion list, so I choose to bother
>you
> >directly.
>
>It is pretty hard to give you a lot of help from the above information.
>It
>is not true to say that read.maimages consistently gives this error. The
>
>above output shows that read.maimages read your first two files
>"EP3_05_ET5_05_0808.spot" and "EP3_28_ET5_28_0908.spot" successfully but
>
>then fails on "SP3_30_EP5_28_0908.spot". What is different about this
>third
>file compared the first two?
>
>Also, what version of limma are you using?
>
>Gordon
>
>
>
> >TIA
> >
> >Christian