[BioC] How to objectively evaluate chip quality?
Nathan.Watson-Haigh at csiro.au
Nathan.Watson-Haigh at csiro.au
Wed Jul 16 01:20:44 CEST 2008
Affymetrix provide several different metrics which can be utilised to
see if your arrays are of good enough quality. However, the actual
cut-off used is always subjective and can only be used as a guide.
Generally speaking, you spend quite a good proportion of your microarray
analysis doing "Quality control through Data Exploration". As such, it's
quite a subjective thing, and you need to produce and explore lots of
different graphs etc in order to get a good picture about the quality of
your arrays. Again, in general, you shouldn't base your opinion about
the quality of an array on a single metric, but use several to inform
you about the quality.
Some specifics about the plots which I use routinely...
1) I use affyPLM to plot pseudo-images of the arrays with the "weights".
This will help you to visualise if any arrays are odd-ones out and have
poor hybridisation due to bubbles on the chip etc. See
http://plmimagegallery.bmbolstad.com/ for examples of really bad chips.
2) I use the "border elements plot" of the AffyQCReport (or a version
I've altered) - again, helps to visualise how consistent hybridisation
is around the edges of the arrays
3) The RNA degradation plot AffyRNAdeg() from the affy package
4) The Affymetrix quality control plot from qc() of the simpleaffy
5) The spike-in control probes table produced by spikeInProbes() from
the simpleaffy package
6) An Eisen plot produced by the made4 package.
7) A PCA plot produced by plotPCA() and a scree plot from the
8) A NUSE plot and RLE plot produced by the affyPLM package
9) A MAD plot produced by affyQAReport() of the affyQCReport package
10) A plot of the SD against the ranked mean intensity of probes using
meanSdPlot() from the vsn package
11) A density plot of the PM probes using plotDensity.AffyBatch() of the
12) Boxplots of the PM probes using boxplot()
I do all the above for raw data and then I do the normalisation and
repeat plots 2, 6, 7, 9, 11 and 12. Then I calculate the gene expression
summaries and use limma to get differentially expressed genes. I use
heatplot() from the made4 package to create heat plots of the
Differentially expressed genes.
So you can see I do a lot of diagnostic/QC/QA plots to explore the data
and to help inform me as to whether any of the arrays should be thrown
out. Be careful not to throw out data just because it doesn't sit well
with your expectations, you need to be able to justify why any array is
discarded, and simply saying that it's an outlier in just one metric is
not usually good enough.
Here's a useful link:
Hope this helps,
From: bioconductor-bounces at stat.math.ethz.ch
[mailto:bioconductor-bounces at stat.math.ethz.ch] On Behalf Of simona
Sent: Wednesday, 16 July 2008 3:35 AM
To: bioconductor mailing list
Subject: [BioC] How to objectively evaluate chip quality?
I have a question about quality assessment of microarray chips. Can
suggest me a quantitative metric to evaluate the chips and in particular
plot about the quality of the chip (for example boxplot of intensity)?
[[alternative HTML version deleted]]
Bioconductor mailing list
Bioconductor at stat.math.ethz.ch
Search the archives:
More information about the Bioconductor