[BioC] beadarray: outliers but not summary in log2-scale?

[Cei Abreu-Goodger]

Hi Mark,

Thanks for your suggestions. I think I will be using the 
method="illumina" for now then.

I've also decided to do vst + rsn, for the simple reason that after an 
vst/rsn normalization I can use the inverseVST function in the lumi 
package to get the real "un-logged" expression numbers back in case 
they're needed (for fold-change calculation, for example).

Thanks again,

Cei

Mark Dunning wrote:
> Hi Cei,
> 
> You can summarize on the unlogged scale by using the option log=FALSE
> in createBeadSummaryData. We're still looking into the best
> combination of outlier removal to use with BASH, it could be that the
> Illumina method for removing outliers by a 3 MAD cut-off is still
> best. This is the default option in beadarray (option
> method="Illumina").
> 
> One final thing, If you plan to use VST method you'll need to
> calculate detection scores prior to running VST, this can be done by:-
> 
> Detection(BSData) = calculateDetection(BSData)
> 
> The type of background correction we don't recommend is the background
> subtraction performed by BeadStudio that subtracts the negative
> controls from each bead-type average on an array. This causes a huge
> increase in variance and missing values if you try and do a
> log-transformation. However, using the combination of VST and quantile
> as you plan should take care of these problems.
> 
> Hope this helps,
> 
> Mark
> 
> On Tue, May 19, 2009 at 2:25 PM, Cei Abreu-Goodger <cei at ebi.ac.uk> wrote:
>> Hi Mark,
>>
>> From reading the createBeadSummaryData help page and the vignettes, I gather
>> that you recommend using BASH to detect outliers, and then using "mean" in
>> log2-space to summarize the data.
>>
>> If I plan to use vst/quantile to normalize my data, what would be the best
>> option for processing the bead level data? I like the idea of summarizing in
>> log2-space but I want a non-log-space result to use with vst. Is there a
>> simple way to accomplish this?
>>
>> And while I'm asking this, I seem to remember hearing that you get better
>> results without using the backgroundCorrect step. Is this recommended?
>>
>> Many thanks,
>>
>> Cei
>>