[BioC] Limma package question

John S. Yordy yordyjs at musc.edu
Tue May 11 20:43:02 CEST 2004

Hi Victor,

The signal mean is used in order to average the signal across the entire
printed spot, thereby taking into account the entire population of
hybridizations for that particular feature on the microarray.  In an ideal
world there would be perfect uniformity in the amount of DNA printed across
the entire feature area, but sometimes this is not the case, and can result
from variation introduced at any step from the printing through the slide
fixation and pre-hybridization.  These variations might manifest as "donuts"
or halos, comets (not extreme -- most software that checks for feature
morphology will throw these out before you get to the normalization step) or
more commonly in my experience, localized "clumping" within the feature.
Because the assumption is that the feature exclusively represents a unique
and specific target for your probe, we sum the intensity of each pixel
within the feature and divide by the number of pixels, thereby arriving at
the signal mean for the feature, which should be more representative of the
total hybridization events than the median.

The background is treated differently because the assumption here is that
there should be no DNA present in the areas between the features, and
therefore there should be no specific hybridization taking place.  However,
again because of random variation that can be introduced throughout the
handling of the slide over the entire course of the experiment, there may be
isolated pockets of increased non-specific background at various points on
the slide.  Generally, because the number of pixels in a given spot's
background having an elevated non-specific signal is relatively few compared
to the total number of background pixels, if one were to take the background
mean, these spurious non-specific signals might contribute to a (falsely)
higher background level than if one would simply take the median intensity
for all the pixels represented in the background and should therefore be
more representative of the actual background signal.


-----Original Message-----
From: bioconductor-bounces at stat.math.ethz.ch
[mailto:bioconductor-bounces at stat.math.ethz.ch] On Behalf Of V.C.L. de Jager
Sent: Tuesday, May 11, 2004 9:30 AM
To: bioconductor at stat.math.ethz.ch
Subject: [BioC] Limma package question

I am using the Limma package 1.6.1 to normalize Imagene files.

Reading the source code I was wondering if there is a particular reason to
take the 'Signal Mean' signal values and the 'Background Median' background
values from the Imagene files. Why not for instance use the 'Signal Median'
column for the signal values?

I hope someone can shed a light on these choices so I can explain them to
some of my colleagues


	Victor de Jager

Bioconductor mailing list
Bioconductor at stat.math.ethz.ch

More information about the Bioconductor mailing list