[BioC] One-Color Agilent miRNA microarrays

[Gordon K Smyth]

On Fri, August 3, 2007 2:40 am, Francesco Favero wrote:
> 2007/8/2, Gordon K Smyth <smyth at wehi.edu.au>:
>>
>> > Date: Mon, 30 Jul 2007 12:39:19 +0200
>> > From: "Francesco Favero" <favero.francesco at gmail.com>
>> > Subject: [BioC] One-Color Agilent miRNA microarrays
>>
>> ...
>> > fit <- lmFit(MA,design,ndups=20,spacing=1,correlation=dupcor$consensus)
>> >> Errore in chol(V) : il minore principale di ordine 2 non è definito
>> >> positivo
>> >
>> > (Sorry for the error in Italian...anyway...)
>> > It works if I don't use ndups, but I need this...
>>
>> What sort of object is MA?  Is it just a matrix, or is it a MAList?
>
> It's just a list of log2 intensities even if is the output of
> normalizeBetweenArrays function.

I suspect you mean that it is a 'matrix'.  It cannot be of class 'list'.  Try

  class(MA)

> In any data analysis, you should check you results at every step.  For
>> example, have you looked at
>> your normalized data?  What is the value of corfit$consensus?
>
> Yes,  I have  a corfit$consensus of 1.

This means your data is degenerate in some way, so you cannot use duplicateCorrelation() without
reworking your data.

> I've seen it's for this I can't use
> lmFit.
> I use ndups=20 so I expect to have some differences between different spot
> for the same genes.. in fact atanh.correlations is good for some gene, but
> for a lot is Inf or NA, maybe this is because only few spots have a good
> Intensity, in average the intensity is very low. this is a problem for miRNA
> in cell-lines. Anyway I think it' possible I still have to work on
> RG$weigths.
>
> Is it so wrong go around all those problems using lm.series function?

Using lm.series is the same as using correlation=0 to lmFit().  Since your estimated correlation
is large, it would probably be better to average over your duplicates.  This can be done by using
avedups() before using lmFit().

Best wishes
Gordon