[BioC] dye swap control: discrepancy of correlation between R vs Rand M vs M

[Peder Worning]

Dear Christine,

We are also using Exiqon miRNA arrays, as you probably would have guessed from my email address. We have very good experience using the Cy3 channel as one channel data and use quantile normalization. For almost every classification or clustering analysis of our data we get better results with the one channel data than with the two channel data. We still have a common ref RNA in the Cy5 channel for comparison and routinely check our results with two channel method as well. 

The one channel analysis require that you are very careful doing the experiments in a standardized way. In fact all array experiments require that.

The way we are checking for dye bias is to repeat one (or more) of the samples as a dye-swap experiment and compare. We get very good reproducibility that way but we also have an ozone free hybridization lab.   

There is none of those problems if you are doing one channel analysis.

I hope it may help you.

/Peder

Best regards 

Exiqon A/S

 Peder Worning, Ph.D.

Senior Scientist, Biomarker Discovery

-----Original Message-----
From: bioconductor-bounces at stat.math.ethz.ch [mailto:bioconductor-bounces at stat.math.ethz.ch] On Behalf Of Christine Voellenkle
Sent: Monday, March 16, 2009 1:04 PM
To: bioconductor at stat.math.ethz.ch
Subject: [BioC] dye swap control: discrepancy of correlation between R vs Rand M vs M

Dear BioC list,
I am working with Exiqon miRNA arrays, performing 2-color labeling,
comparing a cell line control versus cel line treatment
I am trying to find a method to control for labeling bias. I read postings
concerning this topic in the BioC archives, it says that the M values should
be negatively correlated. If they are not, I found the advice to go back to
the R and G intensities and check if they show the exepcted strongly
negative correlation.
I did both and saw the following:
R vs R as well as G vs G strongly positive: 0.94 (not normalized)
 M vs M  negatively correlated (-0.344 not normalized; -0.46 loess
normalized).

I do not understand how this is possible, did I do a mess in the script? If
there is a problem with dye degradation, I should see it also in the M vs M
correlation  as a positve correlation.
Is there any another way to check for labeling bias?
Thanks in advance for your time!
Christine

Please find enclosed the script:
  library("limma")
short_RG<-read.maimages(file="C:/Christine/CV 67.gpr",
source="genepix.median",
other.columns="Flags")

 ds_short_RG<-read.maimages(file="C:/Christine/CV 67_DyeSwap.gpr",
source="genepix.median",
other.columns="Flags")
   x= short_RG$R
y= ds_short_RG$R
cor(x,y)
 #### correlation 0.942051
plot (x=x, y=y)

 x= short_RG$G
y= ds_short_RG$G
cor(x,y)
 #### correlation 0.9397048
plot (x=x, y=y)


 # calculating M values, WIthout normalization
short_RG_M <- normalizeWithinArrays(short_RG, method="none",bc.method =
"normexp", offset = 10)
ds_short_RG_M <- normalizeWithinArrays(ds_short_RG, method="none",bc.method
= "normexp", offset = 10)

 x= short_RG_M $M
y= ds_short_RG_M$M
cor(x,y)
 #### correlation  -0.3445266
plot (x=x, y=y)


  # one-stop normexp bg correction and loess normalization
RG_loessB <- normalizeWithinArrays(short_RG, method="loess", bc.method =
"normexp", offset = 10)
ds_RG_loessB <- normalizeWithinArrays(ds_short_RG, method="loess", bc.method
= "normexp", offset = 10)
 x= RG_loessB$M
y= ds_RG_loessB$M
cor(x,y)
 #### correlation    -0.4643207
plot (x=x, y=y)

-- 
Dr. Christine Völlenkle, Ph.D.
Research Laboratories-Molecular Cardiology
I.R.C.C.S. Policlinico San Donato
Via R. Morandi, 30
20097  S. Donato M.se (MI) Italy
Phone: +39 02 52774 683 (lab)
          +39 02 52774 533 (office)
Fax:    +39 02 52774 666
email: christine.voellenkle at gmail.com

	[[alternative HTML version deleted]]