[BioC] Study design in DEXSeq

[Antonio Domingues]

Hi Alejandro,

Thank you for the quick reply.

I've already done what you've suggested but thought that there might be 
a way of using the experimental condition "total" as a mask (control). 
Meaning that the DEXSeq output (hits) would be the  exons that show 
changes upon knockdown *only* in the fraction. Since that is not 
possible I'll have to stick with the direct comparison of FC.

Cheers,
António


On 04.07.2013 11:34, Alejandro Reyes wrote:
> Dear Antonio Domingues,
>
> It is not possible to modify the DEXSeq formulas in order to test for 
> 'not changes in exon usage'.  An option would be to subset your 
> ExonCountSet object leaving only the subcellular fractions or the 
> totals and do the vanilla DEXSeq analysis in both subsets separately. 
> Afterwards you could compare the results by plotting the fold changes 
> and try to identify abrupt changes in the knockdown effect differences 
> between the cellular fractions and in the total.
>
> Alejandro
>
>> ** I have sent this message before, but somethign must have gone wrong
>> because it seems like it never reached the mailing-list. If it did and
>> did his a duplicate, my apologies **
>>
>> Dear Bioconductors,
>>
>> I would like to ask for some advice/suggestions on the set-up of DEXSeq
>> with multiple condictions. At the moment, I am using DEXSeq in a
>> "vanilla" fashion:
>> - 2 conditions, knockdown and control
>> - 2 biological replicates per condition
>> - output exons that change upon knockdown.
>>
>> So far this is working fine. But I also have another experimental
>> variable: sub-cellular fractions (total vs fraction). The goal is obtain
>> exons whose expression is changed in the knockdown but only in the
>> fraction, that is a combined effect of knockdown and sub-cellular
>> localization. Following the vignette, I was thinking of an experimental
>> design like this:
>>               condition      type
>> sample1_a    control      total
>> sample1_b    control      total
>> sample2_a    knockdown    total
>> sample2_b    knockdown total
>> sample3_a    control      fraction
>> sample3_b    control      fraction
>> sample4_a    knockdown    fraction
>> sample4_b    knockdown    fraction
>>
>> and the code would be:
>> formuladispersion <- count ~ sample + ( condition + type ) * exon
>> ecs <- estimateDispersions( ecs, formula = formuladispersion )
>> ecs <- fitDispersionFunction(ecs)
>> formula0 <- count ~ sample + type * exon + condition
>> formula1 <- count ~ sample + type * exon + condition * I(exon == exonID)
>> ecs <- testForDEU( ecs, formula0=formula0, formula1=formula1 )
>> res2 <- DEUresultTable( ecs )
>>
>> would this work and is this design correct?
>>
>> Thank you,
>> António
>>
>>
>>
>> _______________________________________________
>> Bioconductor mailing list
>> Bioconductor at r-project.org
>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>> Search the archives: 
>> http://news.gmane.org/gmane.science.biology.informatics.conductor
>>
>


-- 
-- 
António Miguel de Jesus Domingues, PhD
Neugebauer group
Max Planck Institute of Molecular Cell Biology and Genetics, Dresden
Pfotenhauerstrasse 108
01307 Dresden
Germany

e-mail: domingue at mpi-cbg.de
tel. +49 351 210 2481
The Unbearable Lightness of Molecular Biology