[BioC] What should happen to control probe information in beadlevel Illumina analyses?

[Simon Lin]

Hi Alice,

Reasonable to exclude them from analysis. In a hypothesis-testing framework, 
you can ignore all the control probes, because we are not interested in 
their differential expression at all.

If you choose to include them, it is an empirical way to empirical your 
false discovery rate, because we know that they should not show up.

Simon

----- Original Message ----- 
From: "Johnstone, Alice" <Alice.Johnstone at esr.cri.nz>
To: "Simon Lin" <simonlin at duke.edu>; "bioconductor" 
<bioconductor at stat.math.ethz.ch>
Sent: Sunday, March 09, 2008 8:56 PM
Subject: RE: [BioC] What should happen to control probe information in 
beadlevel Illumina analyses?


Thanks for your reply Simon,
Yes the expression level is at the low end of the range so I do not
think that this is a hybridisation problem as the original control plots
looked fine.
I used the VST transformation with a quantile normalisation of the Bead
Summary data.
My question is whether it is correct to remove the control probe
information from the Bead Summary object before fitting the regression
model so I don't get any spurious results like I have found, and what is
the best way to do this?

The expression values for one of the control genes are below, the
trt1-control contrast comes up in my topTable results.  This only occurs
when I do my own bead-level summation, the BeadStudio produced bead
summary data does not have the intensity info for this probe.
Trt 1 Control Trt 2
7.544901 7.638846 7.531172
7.515447 7.614039 7.581452
7.495771 7.605357 7.555899
7.485725 7.622735 7.588895
7.538404 7.592215 7.516454

Regards
Alice