[BioC] Limma: Error in backsolve

[Gerhard Thallinger]

Dear all,

  I am analyzing an experiment consisting of 4 one-channel
hybridizations with two conditions and a technical replicate
for each of them. I am using limma 2.4.4 (2005/11/29) under
R 2.2.0 on Windows. The design matrix is as follows:

   0h  2h
1   1   0
2   1   0
3   0   1
4   0   1

Since there are technical replicates the correlation between them
should be taken into account:

  techrep <- c(1,1,2,2)
  corfit <- duplicateCorrelation(signal(anaexp), ndups = 1, block = techrep)
  fit <- lmFit(signal(anaexp), design, block = techrep, cor = corfit$consenus)

The fit aborts with the following error: 
  Error in backsolve(cholV, y, transpose = TRUE) : 
         NA/NaN/Inf in foreign function call (arg 1)
  In addition: Warning messages:
  1: NaNs produced in: sqrt(dfitted.values) 
  2: NaNs produced in: sqrt(dfitted.values) 

The error is probably a consequence of the warnings from the "corfit <-" step.
What value would one expect for the consensus correlation ?
Any ideas how to work around these error ?

The fit without block and cor works flawlessly. Trying to define contrasts
I encountered the following error messages: 

  cont <- makeContrasts("2h-0h", levels = design)
  Error in parse(file, n, text, prompt) : syntax error in "2h"

Obviously a digit as the first character is not allowed. Prepending a "T"
before the condition solved this problem. Using a variable for the contrast
lead to a different message:

  ctr <- "T2h-T0h"
  cont <- makeContrasts(ctr, levels = design)
  Error in eval(expr, envir, enclos) : object "ctr" not found

This could be circumvented by using the following construct:

  cmd <- paste("cont <- makeContrasts(", ctr, ", levels = design)", sep = '"')
  eval(parse(text = cmd))

TIA for your help

Gerhard