Hundreds Of Bad Annotations On Affymetrix Mogene 1.0. Why?

Question

[David Quigley]

This is partly a question, partly a warning.

I'm using the Affymetrix Mouse 1.0 ST microarray annotation NA32, but this may apply to other Affymetrix annotations. On NA32 there are several hundred genes with multiple annotations on different chromosomes that appear to be totally bogus assignments to the gene symbol. Example: Akt2 is on mouse chromosome 7. NA32 has two transcript clusters labeled Akt2: 10551393 and 10374364. The first is on 7, but the second is on 11. Wrong. A lot of these bogus annotations have the word "recovered" in the Affymetrix annotation on Netaffx. Need I say that this has wasted several hours of my life so far?

There are another 500+ genes like this where the annotation may be telling you the correct hybridization location for the probesets in the cluster, but the actual gene assignment seems to be nonsense. Can anyone explain this? I assume the "recovered" word is the key.

[David Quigley]

Thanks. As far as I can see the probes align to the "other" chromosome as advertised; the problem is that those probesets are tagged with a bogus gene name.

[brentp]

You might try to get the annotations from ensembl. I think they re-map the probes based on sequence.

Answer 1

[Adrian]

It looks like the chromosome 11 locus has a pseudogene with considerable identity to Akt2. Presumably (I haven't double-checked in detail) the probeset in question is measuring a sequence that is actually identical between the true Akt and the pseudogene.

[David Quigley]

It's true that there is an Akt2 psuedogene on 11; the symbol is Akt2-ps, Entrez ID 111116. There's at least one paper identifying it. So in this case the annotation should be "Akt2-ps" not "Akt2". It's possible many of the errors are pseudogenes, but that's still bad annotation.

[Adrian]

If you blat the probes in probeset 10374364, 24 / 25 of them are 100% identical to both loci. So there is effectively no distinguishing between them -- they must have originally designed the probeset against the Chr11 locus, and then realized that it also matched to the real Akt, hence "recovered" (If that pseudogene is actually transcribed, whether that measured Akt transcript expression is really the functional Akt might be a concern, but the annotation makes sense.) I'll bet that there is similar redundancy with the the other suspicious probesets, but it would be interesting to check.