[BioC] AffyPLM can detect poor labelling on test chips

[Matthew Hannah]

Hi,

Considering the amount of questions I've asked, I instead thought I'd
post about something that may be of general use/interest.

We had a chip that was an outlier in comparison to the rest of our
dataset (hist(affybatch), RNAdeg, AffyPLM). All our samples are
hybridised to test chips before hybridisation to ATH1 arrays. This
sample passed the Affymetrix QC based on the test chip 3:5 ratios and
scaling factor (although this was slightly higher than normal).

In the post-mortem I decided to fit a PLM model to the test chips and
the same sample was also an outlier (very green in the image plots).
This is despite the fact that only a small proportion of the test chip
probesets are against arabidopsis (clearly the NS binding was
sufficient). This helped to clearly identify that the labelling had
failed, rather than it was just a bad ATH1 chip.

Also as the test .CEL files are very small the PLM only takes a few
minutes, even with lots of them.

Also of interest was that rather than just looking at 3:5 ratios of the
test chip control genes, you can also do a simple xy scatter plot (just
affy MAS5 signal values) of the control spots for related samples and in
our case there was usually a resonable correlation. The outlier was also
obvious using this method.

If anyone has a poor affy array-test chip pair, I'd be interested if
these methods also work - let me know.

If anyone's still using test chips, then perhaps this might save someone
a wasted chip.

Cheers,

Matt