[BioC] Microarray quality control question

[Naomi Altman]

Dear Lee Ann,

I prefer to answer on the list so I hope that you do not mind that I 
am forwarding my response.

In a dye-swap experiment, there should be a negative correlation 
between the dye-swaps.  If there is not, this is a serious problem - 
either there is a problem with the dyes, or the slides were set up 
with 2 WT or 2 MU on the same slide.  Your design seems to indicate 
that you used a reference design - i.e. 10 slides with sample and 
reference on each.  Is that right.

  I am not sure what you mean by "genes on all the slides seem to be 
down-regulated".  If this means that R<G on all slides, then this 
would seem to indicate a problem with the dye.  This can be due to 
ozone degradation, which appears to affect red more.  The bimodal 
peak might also indicate this, however, I have often seen bimodal 
distributions, which might be due to a set of oligos which do not 
hybridize (either RNA absent or poor chemistry).  It could be the RNA 
creating the bimodality, but that does not explain the positive 
correlation.  If down-regulated means "compared to the reference" 
then it seems like that would create a negative correlation.

I think you set up the ANOVA correctly.  My experience with poor dye 
suggests that including a dye effect does not adequately handle the problem.

Maybe you should post of couple of these plots to the e-mail list and 
see what others think.

--Naomi


At 02:51 PM 1/20/2006, you wrote:
>Dr. Altman,
>
>I have taken special note and followed all your suggestions 
>pertaining to problems similar to mine.  The experiment we ran was 
>with 5 biological replicates of 2 different strains of 
>mice.  (WT1-WT5 and MU1-MU5)  As well, each pair had a dye-swap 
>(technical replicate).   I used  loess normalization for within 
>Array and Quantile normalization for between Array.
>My issues of concern are that the my dye swapped slides  show 
>positive correlation between them. (The genes on all the slides seem 
>to be heavily down regulated)  As you advised on one of the message 
>boards I plotted the M vs M values for the swapped slides and two 
>sets of them show a very dramatic positive correlation.  Does this 
>warrant that these slides should be redone or can be analyzed as 
>is.  The biologists in the lab have expressed to me that this is 
>in-house RNA and the quality may not be as high as company RNA.
>Is this much dye effect truly accounted for with the right design, 
>(I used design<-  (DyeEffect =1, c(rep(1,-1),5)).   I included 
>blocking for the biorep pairs.  My list of significant genes for 
>DyeEffect is around 6000.  With this question of dye bias is it 
>completely wrong to just use a design<- c(rep(1,-1),5)?
>
>Another plot that is confusing to me is the density plots.  On 
>almost half of the slides there is a bimodal peak for the density 
>plot of the oligos.  From what I have read this may be the cause of 
>a spatial effect on the slide.  After looking at all the image plots 
>I do see a streak in the middle of each array and this was caused by 
>the printing machine missing a row of spots.  I accounted for the 
>missing spots in the analysis.  Is this bimodal histogram a sign of 
>something else and is there anything I can do to accomodate for it?
>
>Lee Ann

Naomi S. Altman                                814-865-3791 (voice)
Associate Professor
Dept. of Statistics                              814-863-7114 (fax)
Penn State University                         814-865-1348 (Statistics)
University Park, PA 16802-2111