[BioC] Paired t-test query

[Gordon K Smyth]

Dear Priyanka,

I think you are making the analysis far complex than it needs to be. 
Your data seems to be the simplest possible two colour design, making 
direct comparisons of two treatments, with all arrays biologically 
independent.  The data example you give would be analysed perfectly well 
using the advice in the limma User's Guide in Section 8.1.  There is no 
need to separate the channels, and there is no paired structure with the 
samples that introduces correlations between arrays, so no need for a 
paired t-test.

Having said that, the separate channel analysis you give would appear to 
be correct if you removed the Pairing variable from the model matrix. 
There is no pairing in your data, other than the pairing within arrays, 
which is already taken into account by the limma separate channel 
analysis.

Best wishes
Gordon


> Date: Thu, 6 Jan 2011 11:34:46 -0600 (CST)
> From: "Kachroo, Priyanka" <priya_coll at neo.tamu.edu>
> To: bioconductor at r-project.org
> Subject: [BioC] Paired t-test query
> Message-ID: <1965222321.935221294335286256.JavaMail.root at neo-mail-3>
> Content-Type: text/plain; charset=utf-8
>
> Hi All,
>
> I would like to ask the forum if a paired t-test is better suited for my 
> experimental design or a time course analysis. I have three time points 
> T0, T6 and T12 for a group of animals. I need to evaluate the DE genes 
> between T6&T0 and also T12&T0. Since the same set of animals were 
> involved at all three time points, will a paired t-test for T6-T0 and 
> T12-T0 be a better strategy or a time course analysis.I have used dual 
> color arrays and hence separated the channels before fitting the model.
>
> For example when i try to do a paired t-test between T0 and T6 i get 50 
> warning messages. I have 5 different animals (biological replicates) 
> that were given no treatment at T0 and a treatment at T6. I tried to run 
> the code as given in limma manual for paired t-test. Since i have two 
> color array i separated the channels first. However i do not understand 
> where the code is wrong. Therefor can someone please explain if the code 
> is right for performing a paired t-test and whether performing a t-test 
> at all is a good idea or not for my design.
>>
>> targets<-readTargets("targets.txt")
>>> RG<-read.maimages(targets,source="genepix",columns=list(R="F635 Median",G="F532 Median",Rb="B635",Gb="B532"))
>> Read 14117071.gpr
>> Read 14117070.gpr
>> Read 14116987.gpr
>> Read 14117067.gpr
>> Read 14117099.gpr
>>> RG$genes<-readGAL()
>>> RG <- backgroundCorrect(RG, method="normexp", offset=50)
>> Green channel
>> Corrected array 1
>> Corrected array 2
>> Corrected array 3
>> Corrected array 4
>> Corrected array 5
>> Red channel
>> Corrected array 1
>> Corrected array 2
>> Corrected array 3
>> Corrected array 4
>> Corrected array 5
>>> MA.p <- normalizeWithinArrays(RG)
>>> MA.Aq<-normalizeBetweenArrays(MA.p,method="Aquantile")
>>> targets2<-targetsA2C(targets)
>>> targets2
>>    channel.col SlideNumber     FileName Identity Pairing Target
>> 1.1           1    14117071 14117071.gpr       61       1     T6
>> 1.2           2    14117071 14117071.gpr       61       1     T0
>> 2.1           1    14117070 14117070.gpr      123       2     T6
>> 2.2           2    14117070 14117070.gpr      123       2     T0
>> 3.1           1    14116987 14116987.gpr      308       3     T6
>> 3.2           2    14116987 14116987.gpr      308       3     T0
>> 4.1           1    14117067 14117067.gpr      315       4     T0
>> 4.2           2    14117067 14117067.gpr      315       4     T6
>> 5.1           1    14117099 14117099.gpr      319       5     T0
>> 5.2           2    14117099 14117099.gpr      319       5     T6
>>> u<-unique(targets2$Target)
>>> Pairing<-factor(targets2$Pairing)
>>> Exposure<-factor(targets2$Target,levels=c("T0","T6"))
>>> design<-model.matrix(~Pairing+Exposure)
>>> corfit<-intraspotCorrelation(MA.Aq,design)
>> There were 50 or more warnings (use warnings() to see the first 50)
>>
>
> Priyanka Kachroo
> Graduate Assistant Research
> Texas A&M University

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}