[BioC] Nested Design (Again) & Subset WithinArray Correlation

[Y. Osee Sanogo]

Dear Gordon,

Thank you for the code. It works!!
My only question is does it then matter that the probes set are not
duplicated and there is no technical replicate per se? I thought
duplicateCorrelation is meant for duplicates or technical replicates? Please
clarify.

Thanks again.

Osee


On 7/28/10 6:53 PM, "Gordon K Smyth" <smyth at wehi.EDU.AU> wrote:

> Dear Osee,
> 
> I haven't seen anyone else try to answer your first question, so I will.
> 
> You're trying to put too many terms in your design matrix, making the
> experiment much more complicated than it actually is.  Your experiment
> simply compares two treatment groups.  It doesn't make sense to estimate
> effects for fish or tanks, because these are just your randomly sampled
> experimental units.  The only real complication of your experiment is that
> some fish share the same tank, so you need to allow for possible
> correlations with a tank.  You can do this is limma by:
> 
>    design <- model.matrix(~Key)
>    fitcor <- duplicateCorrelation(ES,design,block=tank)
>    fit <- lmFit(ES,design,block=tank,correlation=fitcor$consensus)
>    fit <- eBayes(fit)
>    topTable(fit,coef=2)
> 
> This approach finds genes which respond to your treatment.
> 
> Best wishes
> Gordon
> 
>> Date: Tue, 27 Jul 2010 06:57:36 -0500 (CDT)
>> From: "Y. Osee Sanogo" <sanogo at illinois.edu>
>> To: bioconductor at stat.math.ethz.ch
>> Subject: [BioC] Nested Design (Again) & Subset WithinArray Correlation
>> 
>> Hello,
>> 
>> I have two questions which may be really trivial...but since I am stuck,
>> I'll appreciate any help.
>> 
>> Question 1: Nested design: This has been addressed before, but I am just not
>> sure whether I am doing it right. The experiment consisted of two groups of
>> fishes (treated and not treated) with three tanks in each group. Each tank
>> hosted three fishes (total =18) of those fishes n=10 (5 per treatment group)
>> were selected for microarray (Notice unequal number of fishes per tank!).
>> 
>> I am interested in 1) Treatment effect (individual fishes)
>>                                2) Treatment effect (fishes nested within
>> tanks, i.e. Need to average the gene expression of fishes within each tank )
>>                                3) Whether there is tank effect
>> 
>> #ExpressionSet =ES_Filt
>> #targets= see below:
>> 
>>      Sample       Key   tank Fish         SAMPLE_LABEL
>> 25407102_532.xys     CON 1 CON_3 SOM01K28
>> 25407202_532.xys     CON 1 CON_2 SOM01K29
>> 25414902_532.xys     EXP 2 EXP_1 SOM01K2D
>> 25407302_532.xys     CON 3 CON_1 SOM01K2C
>> 25406602_532.xys     EXP 4 EXP_2 SOM01K25
>> 25407002_532.xys     EXP      4 EXP_3 SOM01K27
>> 25415502_532.xys     EXP 4 EXP_4 SOM01K2E
>> 25405602_532.xys     CON 5 CON_4 SOM01K23
>> 25406702_532.xys     CON 5 CON_5 SOM01K26
>> 25415702_532.xys     EXP 6 EXP_5 SOM01K24
>> 
>> I have tried the following design based upon what I found online, but was
>> not really sure whether this is the right way of doing it.
>> 
>> design.nested_ES<- model.matrix(~Key + (tank/Fish), data=targets)
>> colnames(design.nested_ES)
>> #I am getting many contrasts, and I am not sure which one represents
>> ?tank/Fish?
>> 
>> fit.nested_ES <- lmFit(ES_Filt, design.nested_ES)
>> Fit.nested_ES <- eBayes(fit.nested_ES)
>> Pred2_Nested_ES<-topTable(Fit.nested_ES, coef=2, adjust="BH", n=Inf)
>> Pred2_Nested_ES[1:10,]
>> 
>> I will really appreciate your help.
> 
> ______________________________________________________________________
> The information in this email is confidential and inte...{{dropped:7}}