[BioC] Duplicate Correlation through Blocks

[Francois Pepin]

Hi Scott,

I replied on this topic earlier today.

>From what we've seen looking at the Agilent chips from our labs (mice
and human, 1x44 and 4x44 chips, made by several people), duplicate
probes on a chip give more or less identical results.

I'll let others say if it's appropriate to use blocking variable that
way, but I do not think the result would be very meaningful. It would
give you an estimate of a intra-chip technical variation that is
basically negligible compared to inter-chip technical variation.

Dealing with those duplicate probes "properly" would be a lot of work,
and I think it would give you very little at the end. We look at them as
a control that the signal from a given probe is reproducible and then
ignore them afterward.

Francois

On Mon, 2007-07-16 at 16:23 -0500, Franklin, Scott wrote:
> I am very new to microarray experiments and have a question with
> calculating
> duplicate correlations for 4x44 Agilent chips.
> 
> After searching extensively through the bioconductor mailing list
> submissions, I have seen that no one has come up with a solution to
> handling
> a variable number of gene replicates on the same chip.  As I understand
> it,
> limma was not designed to handle this.
> 
> Before seeing that this was not a functionality of limma, we attempted
> to
> treat each individual gene as a block.  We sorted by probe name, then
> assigned a unique number to each probe name with it repeating the number
> of
> times that gene appears on the chip.  We ended up with roughly 21,000
> blocks.  I had assumed that duplicate correlation would compute a
> correlation for each of these blocks.
> 
> Is this a reasonable approach or a complete misunderstanding of the
> blocking
> variable? (By the way, the calculation is still running after an hour)
>