array quality control, affy probes, contracts setting in limma, result interpretation and weights
1
0
Entering edit mode
alakatos ▴ 110
@alakatos-6983
Last seen 2.1 years ago
United States

Hello,

I used oligo and limma packages to process and analyze my Affymetrix Mouse Gene 2.0 ST Arrays.

The ExpressionSet  object was created by:

eSet = rma(raw, target="core")

eSet:
ExpressionSet (storageMode: lockedEnvironment)
assayData: 41345 features, 19 samples
element names: exprs
protocolData etc......

After RMA, the boxplot and array clustering demonstrated good quality.  No outlier.

My design matrix:

design <- model.matrix(~0+Treatment)  #4 level - treatment with subsequent contrast matrix.

Question 1 :

In the topTable, the top significant results were annotated as "reporter probe", for instance "17549014" (category: reporter).

I am wondering what could be the explanation.  Could it be  the consequence  of poor array quality? If it is, is there any solution for additional correction besides RMA? What about array weights?

I understand I could filter my probes and keep "main" ones but still I would like to know the meaning of my result with respect to array quality without probe filtering.

Question 2:

In order to detect DE  I could choose between 2 approaches. One approach is  to set all the treatments (3)  against the control group and overlap the results (venn diagram). The other one is simple to set all the possible contrasts.  I am wondering if there is any difference between the 2 approaches. If it is , under what criteria should I prefer one to the other?

Anita

0
Entering edit mode

I would be interested to know the average expression levels for the reporter probeset that is coming up in your topTable() results. This probeset doesn't have any matches in the mouse genome, so it is unlikely to be measuring any transcript. However, Rafa showed long ago that you can get Affy probes to light up even if you don't even hybridize any cDNA to the array, so it might just be low level noise, which can be really problematic if you don't filter out the really low expressing probesets.

1
Entering edit mode
@gordon-smyth
Last seen 1 hour ago
WEHI, Melbourne, Australia

With respect your Question 2, deciding what contrasts to make between the treatments depends on what biological questions you have. If you are interested in the difference between each of the three treatments and controls, then you should naturally form the contrast of each treatment vs the control group. If you are interested in comparing the treatments directly with each other, then you should also make a direct contrasts between them. Deciding which contrasts are of interest to you depends on your own biological and scientific questions, and of course only you know what these are.

With respect to Question 1, I don't know what "reporter" probes are on this particular Affy array nor whether there is any reason to expect them to have constant expression between arrays. You would need to go to the NetAffx website for more information on these probe-sets.