[BioC] Comparison of your beadarray package with Illumina's analysis software

[Mark Dunning]

Hi Benno,

I have not done any systematic comparisons of the methods used by  
Illumina and limma to detect DE genes. I'm not sure if anyone else  
has either?

I'm not surprised that there are differences in the methods. In limma  
you are using log2 values whereas Illumina like to keep all values on  
the unlogged scale and use the relationship between the means and  
variances of each bead type in their model.

Illumina use a modified t-statistic, calculate the p-value and then  
do a log10 transformation (plus some other stuff...) to get the  
DiffScore, so this will not be on the same scale as the moderated-t  
statistic or B values from limma.

What statistics are you comparing for each method?

Hope this helps,

Mark


On 2 Jul 2007, at 10:47, Benno Puetz wrote:

> Hi Mark,
>
> I have been using your package (and limma) for analysis when we  
> recently tried to compare it to Illumina's approach.
> One thing we were surprised about were huge differences in the  
> reported significance of the genes detected, also the regulation  
> value differed, sometimes significantly.
>
> Have you done any systematic comparisons and ideas as to how these  
> discrepancies should be handled? Illumina seems to find more genes  
> - are these reliable or is a more conserative approach advised?
>
> Your comments will be appreciated
>
>    Benno