[BioC] AFFX probes as result of differential expression analysis

[James W. MacDonald]

Hi Constanze,

Getting 'AFFX' probes differentially expressed may or may not be a 
problem, depending on the probes themselves, so a blanket description of 
'AFFX' isn't helpful.

For instance, probes of the form AFFX-HUM are actually interrogating the 
product of human genes. For instance, beta-Actin, GAPDH, STAT1, etc. 
While these are assumed to be 'housekeeping' genes, I don't know that 
anybody has shown conclusively that they never change expression from 
sample to sample.

The Dap, Phe, Lys, and Thr are bacterial control genes that are added 
before the in vitro transcription step as a control to test the quality 
of the IVT step. If they are different, it may signal a problem with 
this step. It may also indicate a batch problem, if the chips were run 
in batches.

The bioB, bioC, bioD and Cre probes are hybridization controls that are 
added to the fragmented cDNA prior to the hybridization step, and may 
indicate differences in the efficacy of the hybridization or batch 
problems (e.g., inaccurate pipetting of the controls, differences in the 
hyb oven settings, etc).

If you are getting batch problems that show up in your comparisons, that 
might indicate that the samples weren't randomized to different batches 
(e.g., all the controls were run in one batch, then the experimentals in 
another batch). If that is true then your biological and technical 
variability are confounded, and there is no way to know if any 
differences are due to underlying biology, or to technical artifact.

One other possibility is that the _5_at and possibly the _M_at probes 
are coming up as differentially expressed. All of these probes are 
directed towards the 3' end (the _3_at probes), the middle of the 
transcript (the _M_at probes) and the 5' end (the _5_at probes).

If you are getting lots of 5' or 5' and middle probes from the 
housekeeping genes different in one set of samples vs the other, this 
may indicate differential mRNA degradation of the sample. If the same is 
true of the Dap, Phe, Lys, Thr genes, this would indicate a problem with 
the IVT step for some of the samples. An RNA degradation plot would help 
to see if that is a problem for some samples.

Best,

Jim



Constanze Schmitt wrote:
> Hello,
> I am doing a differential expression analysis on Affy HGU95A and
> HGU95AV2 chips. After combining the data into one batch
> (merge.AffyBatch)and jointly  normalising with rma (subsequent quality
> check was ok), I used the "limma" package to fit a linear model (lmFit)
> and then an empirical Bayes model(eBayes). Results are ranked with
> topTable (fdr correction).
> I get AFFX probes in my top 10 results; as far as I know these control
> probes should not be differentially expressed. Does this hint at
> possible batch-effects in my data even though there are no outlier chips
> in the heatmap, intensity- and boxplots of the normalized chips?
> I'd appreciate any help or ideas,
> thanks,
> Constanze
> 
> 
> 
> 
> 

-- 
James W. MacDonald, M.S.
Biostatistician
Douglas Lab
University of Michigan
Department of Human Genetics
5912 Buhl
1241 E. Catherine St.
Ann Arbor MI 48109-5618
734-615-7826
**********************************************************
Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues