[BioC] method for removing consistent technical bias?

[k. brand]

Dear BioCers,

I have a consistent, reproducible technical discrepancy resulting from 
two different hybridisations. Two biological replicates for Hyb A used 
expired arrays, and has significantly lower intensities than the two 
biological replicates of hyb B, which behave normally. I thus have 4 
biological replicates of three different tissues which cluster (K-means) 
more strongly by hyb. than by tissue.

RMA does a courageous job normalising the discrepancy (see summary of 
normalised and unnormalised data below), but if any one has experience 
or suggestions they care to share on getting the most out of this flawed 
dataset, id be very grateful.

Of note- since the 3 tissues are 'paired', ie come from the same animal, 
i was also considering a paired ANOVA of the 3 tissues, reducing, if not 
eliminating the need to overcome inter-hyb variation, but lack the 
experience to know what/if there is an appropriate R implementation. 
Furthermore i have no idea how to use the 2/4 replicates to increase the 
statistical power with such an approach. Any guidance appreciated.

thanks in advance,

Karl


#Unnormalized data:
  > dat <- ReadAffy()
  > dat.rma <- rma(dat, normalize=FALSE)
Background correcting
Calculating Expression
  > apply(exprs(dat.rma),2,summary)
          Tco1A.CEL Tco2A.CEL Tco3B.CEL Tco4B.CEL Tmi1A.CEL Tmi2A.CEL
Min.        1.633     1.713     1.869     2.431     2.027     1.736
1st Qu.     2.627     2.751     3.234     3.554     2.831     2.882
Median      3.329     3.320     4.933     4.952     3.433     3.588
Mean        4.153     3.902     5.572     5.741     4.330     4.261
3rd Qu.     5.147     4.541     7.534     7.556     5.254     5.151
Max.       14.180    13.640    14.370    14.280    14.240    13.880

          Tmi3B.CEL Tmi4B.CEL Tsh1A.CEL Tsh2A.CEL Tsh3B.CEL Tsh4B.CEL
Min.        1.771     2.575     1.607     1.902     1.771     2.360
1st Qu.     3.280     3.716     2.541     2.850     3.197     3.940
Median      4.986     4.978     3.183     3.357     4.833     5.488
Mean        5.629     5.762     3.983     4.032     5.493     6.092
3rd Qu.     7.606     7.449     4.882     4.629     7.426     7.892
Max.       14.300    14.180    14.070    13.990    14.230    14.340

#RMA normalized data:
  > eset <- justRMA(filenames=list.celfiles())
Background correcting
Normalizing
Calculating Expression
  > apply(exprs(eset),2,summary)
          Tco1A.CEL Tco2A.CEL Tco3B.CEL Tco4B.CEL Tmi1A.CEL Tmi2A.CEL
Min.        1.945     2.077     2.014     2.070     1.970     2.050
1st Qu.     3.394     3.574     3.114     3.135     3.381     3.483
Median      4.469     4.533     4.552     4.435     4.474     4.472
Mean        5.191     5.110     5.221     5.192     5.199     5.133
3rd Qu.     6.539     6.114     6.947     6.866     6.589     6.260
Max.       14.180    14.170    14.120    14.170    14.170    14.180

          Tmi3B.CEL Tmi4B.CEL Tsh1A.CEL Tsh2A.CEL Tsh3B.CEL Tsh4B.CEL
Min.        1.930     2.056     2.015     2.028     1.815     1.919
1st Qu.     3.106     3.190     3.436     3.524     3.132     3.134
Median      4.520     4.426     4.497     4.484     4.548     4.450
Mean        5.219     5.194     5.203     5.140     5.224     5.198
3rd Qu.     6.949     6.824     6.537     6.241     6.938     6.887
Max.       14.130    14.070    14.180    14.130    14.110    14.150

  > sessionInfo()
Version 2.3.0 (2006-04-24)
i386-pc-mingw32

attached base packages:
[1] "tools"     "methods"   "stats"     "graphics"  "grDevices" "utils"
      "datasets"  "base"

other attached packages:
       affyPLM        gcrma  matchprobes     affydata mouse4302cdf
     vsn        limma         affy       affyio      Biobase
       "1.8.0"      "2.4.1"      "1.4.0"      "1.8.0"     "1.12.0"
"1.10.0"      "2.7.3"     "1.10.0"      "1.0.0"     "1.10.0"


-- 
Karl Brand <k.brand at erasmusmc.nl>
Department of Cell Biology and Genetics
Erasmus MC
Dr Molewaterplein 50
3015 GE Rotterdam
lab +31 (0)10 408 7409 fax +31 (0)10 408 9468