[BioC] Limma: NaNs in intraspotCorrelation()

[Muller, Pie]

Dear Gordon and Bioconductor list members,

I am using limma to analyse an unlinked two-colour array experiment consisting of 31 arrays (with 34,992 spots on each array) and 38 unique targets. I would like to separate the two channels to perform a separate channel analysis using the function lmscFit() and tried to follow the example given in the limma user's guide (see below) but found that the "intraspotCorrelation()" gives me nothing but NaNs. What could have gone wrong? There are no NaNs in either the M or A values in my MAlist object. I suspect there might something wrong with my design matrix as I noticed that by reducing the design matrix to one single factor such as "design=matrix(rep(1,62), ncol=1)" I receive a consensus correlation value back. However, there is still another warning message "reml: Max iterations exceeded in: remlscore(y, X, Z)". Does anyone have suggestions as to solve this problem?

In advance, many thanks for your support!

Pie


My target file:

> targets
                     Array     Cy3     Cy5
1  2007-06-07_B2162-09.gpr F.168.C  F.72.B
2  2007-06-07_B2162-13.gpr  U.48.A  U.24.A
3  2007-06-07_B2162-16.gpr  F.24.E  U.24.C
4  2007-06-07_B2162-20.gpr   F.8.C   U.8.C
5  2007-06-07_B2162-22.gpr  U.24.A   U.8.A
6  2007-06-15_B2162-02.gpr  F.48.B  F.24.D
7  2007-06-15_B2162-04.gpr  U.72.A  U.48.A
8  2007-06-15_B2162-05.gpr  F.72.B  F.48.B
9  2007-06-15_B2162-07.gpr  F.48.C  U.24.C
10 2007-06-15_B2162-10.gpr F.168.C   U.8.B
11 2007-06-15_B2162-11.gpr U.168.A  U.72.A
12 2007-06-15_B2162-19.gpr   F.8.A U.168.A
13 2007-06-15_B2162-23.gpr  F.72.C  U.72.C
14 2007-06-15_B2162-25.gpr  F.24.D   F.8.B
15 2007-06-15_B2162-26.gpr F.168.D U.168.C
16 2007-09-26_B2165-05.gpr  I.72.A  I.48.A
17 2007-09-26_B2165-08.gpr  I.24.A   I.8.A
18 2007-09-26_B2165-12.gpr   I.8.A F.168.B
19 2007-09-26_B2165-15.gpr   I.8.A I.168.B
20 2007-09-26_B2165-18.gpr I.168.A  I.72.A
21 2007-09-26_B2165-22.gpr  I.48.A  I.24.A
22 2007-09-28_B2165-02.gpr U.168.C I.168.C
23 2007-09-28_B2165-03.gpr I.168.C F.168.C
24 2007-09-28_B2165-04.gpr  I.24.B  F.24.D
25 2007-09-28_B2165-11.gpr  I.72.B  F.72.B
26 2007-09-28_B2165-14.gpr   U.8.C   I.8.C
27 2007-09-28_B2165-17.gpr  U.72.C  I.72.C
28 2007-09-28_B2165-19.gpr  I.48.B  F.48.B
29 2007-09-28_B2165-20.gpr  U.48.C  I.48.C
30 2007-09-28_B2165-23.gpr   I.8.B   F.8.B
31 2007-09-28_B2165-26.gpr  U.24.C  I.24.C


My code to generate the design matrix for single channel analysis:

# seperate channels
targets.sc=targetsA2C(targets)
u=unique(targets.sc$Target)
f=factor(targets.sc$Target, levels=u)
design=model.matrix(~0+f)
colnames(design)=u


The warnings produced by corfit=intraspotCorrelation(MA, design)

...
Warning message:
NaNs produced in: log(x) 
Warning message:
NaNs produced in: log(x) 
Warning message:
NaNs produced in: log(x) 
Warning message:
NaNs produced in: log(x)

> corfit$consensus
[1] NaN


The warning produced if the design matrix is reduced to one single factor:

> design=matrix(rep(1,62), ncol=1)
> corfit=intraspotCorrelation(MA, design)
Warning message:
reml: Max iterations exceeded in: remlscore(y, X, Z)

-------------------------------------

Dr Pie Müller
Vector Group
Liverpool School of Tropical Medicine
Pembroke Place
Liverpool
L3 5QA
UK

Tel +44(0) 151 705 3225
Fax +44(0) 151 705 3369

http://www.liv.ac.uk/lstm
http://www.ivcc.com