Hello,

A colleague asked me to analyze a 2-dye Agilent experiment available in GEO:

http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE14594

Normalizing etc. goes fine, but I am struggling with defining the design, specifically whether I should block for “mouse” or not. If so, how to do this for a 2-dye experiment? (I am asking/confused because I am most experienced with Affymetrix [single dye] experiments...).

Let me explain: the design is rather straight-forward: 2 cell types have been isolated from 2 mice, all samples have been labelled with both Cy3 and Cy 5 (dye-swap), and samples have been arrayed on in total 4 slides. See target file below.

My aim: I (only) would like to know the differential gene expression between the 2 cell types. Intuitively I feel I can (should?) block on “mouse”, but is this needed/possible? If so, is the way I am doing this OK?

Any advice is appreciated!

Thanks,

Guido

> targets
               FileName             Cy3             Cy5 mouse
GSM364903 GSM364903.txt       SI_crypts crypt_organoids     1
GSM364904 GSM364904.txt crypt_organoids       SI_crypts     1
GSM364905 GSM364905.txt       SI_crypts crypt_organoids     2
GSM364906 GSM364906.txt crypt_organoids       SI_crypts     2
>

# Build design
design <- modelMatrix(targets, ref="SI_crypts")

>Found unique target names:
> crypt_organoids SI_crypts

# add 'mouse' to model 
mouse <- as.factor(targets$mouse)
design <- cbind(design, mouse)

> design
          crypt_organoids mouse
GSM364903               1     1
GSM364904              -1     1
GSM364905               1     2
GSM364906              -1     2
>

# check
> is.fullrank(design)
[1] TRUE
> nonEstimable(design)
NULL
>

fit <- lmFit(MA.q3, design) #Have I properly blocked for mouse?

fit2 <- eBayes(fit, trend=TRUE)
> topTable(fit2)
      Row Col Start                                                     Sequence
469     6  47  6058 CTCATTTCAGGGTTTGATGGTAGACTCAAGAGCACGGACTTCCTTTAGTAAGAACCAGTC
11352 135  15   621 GTGAAGACTCACCTAGAAAAAGCATTTTCTATCTACAGTCCACATTAGAGCCTTAATCTG
41229 488  20  1064 TCTGCAAACTCAGTCTGAACAATTCGCTGTGTCCACACTCTCAGCTCATCGTGTTTTCAA
7523   89  79  1755 AGGAGGGACCAGGTTCAGGGTCTTTCTATGTGGTATCTTCCAAAGTGCAACTTTTCACTT
12393 147  41    60 TTGAGAGCTAGTGCCCGTGTGCCCATTAGGTCCCTTTTAGAAGCTGTGCTTGCCAGCTCC
38002 450  15   409 AACACTGTCCTCCGTTTCATTAGTGGGCAATCTGGCTCTGCCATAGATTCTATTAGCCTG
38340 454  14   913 ACAGAGACAACTATTCTGGAGGAGTCGTCAACTACCACATGAAGGAAGACGGTTGGGTGA
8086   96  48   856 ACCTCAGTCCTGAAGAGGCCTACGACCTTGGCCGCAGAGCTATTGCTTATGCTACCCACA
16152 191  82   802 AAATACAAGAACTGTTTATAGGAGACCAGTTGATCTTGGGAAAACAGCAGTTGTGTGAGC
29257 346  71   546 TGGTTCCGGAATGGCCAGGAGGAGACGGTGGGGGTCTCATCCACACAGCTTATTAGGAAT
      ProbeUID ControlType    ProbeName           GeneName SystematicName
469        335           0 A_52_P164797                Lct   NM_001081078
11352    10497           0 A_51_P169671              Reg3b      NM_011036
41229    37989           0 A_51_P481159               Cbr3      NM_173047
7523      6982           0 A_51_P261737              Pdzk1      NM_021517
12393    11492           0 A_52_P460977 ENSMUSG00000073645       BC025933
38002    35004           0 A_51_P140429      1810010M01Rik      NM_026918
38340    35305           0 A_51_P345366              Psmb8      NM_010724
8086      7396           0 A_51_P345367              Psmb8      NM_010724
16152    15008           0 A_52_P582059               Lyz1      NM_013590
29257    10301           0 A_51_P215237             H2-Ab1      NM_207105
                                                                                                                              Description
469                                                                                   ref|Mus musculus lactase (Lct), mRNA [NM_001081078]
11352                                                        ref|Mus musculus regenerating islet-derived 3 beta (Reg3b), mRNA [NM_011036]
41229                                                                      ref|Mus musculus carbonyl reductase 3 (Cbr3), mRNA [NM_173047]
7523                                                                   ref|Mus musculus PDZ domain containing 1 (Pdzk1), mRNA [NM_021517]
12393                         gb|Mus musculus predicted gene, ENSMUSG00000073645, mRNA (cDNA clone IMAGE:5068496), partial cds [BC025933]
38002                                                       ref|Mus musculus RIKEN cDNA 1810010M01 gene (1810010M01Rik), mRNA [NM_026918]
38340 ref|Mus musculus proteasome (prosome, macropain) subunit, beta type 8 (large multifunctional peptidase 7) (Psmb8), mRNA [NM_010724]
8086  ref|Mus musculus proteasome (prosome, macropain) subunit, beta type 8 (large multifunctional peptidase 7) (Psmb8), mRNA [NM_010724]
16152                                                                                ref|Mus musculus lysozyme 1 (Lyz1), mRNA [NM_013590]
29257                                        ref|Mus musculus histocompatibility 2, class II antigen A, beta 1 (H2-Ab1), mRNA [NM_207105]
      crypt_organoids        mouse  AveExpr         F      P.Value    adj.P.Val
469         -4.520932  0.156136598 13.56408 1305.8464 2.554305e-08 0.0007871481
11352       -4.290591  0.137475574 14.45943 1048.9086 4.764965e-08 0.0007871481
41229        5.186591  0.110893726 11.85187  933.0347 6.647739e-08 0.0007871481
7523        -3.928422 -0.011424487 11.32021  671.0715 1.696315e-07 0.0007871481
12393       -3.271531  0.092322914 11.87628  647.4011 1.878468e-07 0.0007871481
38002       -3.510685  0.154594450 15.00852  633.5930 1.997081e-07 0.0007871481
38340       -3.469233  0.011392574 12.15195  606.4721 2.261245e-07 0.0007871481
8086        -3.314961  0.006260582 12.32464  580.9312 2.555046e-07 0.0007871481
16152       -2.922789  0.152916312 16.21389  553.3243 2.933709e-07 0.0007871481
29257       -3.296891 -0.114447086 10.85748  549.9973 2.984355e-07 0.0007871481
>

No, you haven't blocked for mouse properly -- it isn't correct to cbind a factor to a design matrix as if it was a continuous covariate.

It is probably not necessary to block on mouse here. The two colour design already adjusts for baseline differences between the two mice.

Based on previous experience with Agilent arrays, it is probably more important to allow for probe-specific dye effects, which you can do by:

design <- modelMatrix(targets, ref="SI_crypts")
design <- cbind(Dye=1, design)