[BioC] LIMMA moderated standard errors?

[Paul Harrison]

Hello,

I'm trying to analyze a set of Combimatrix arrays with LIMMA and have
a couple of questions. I haven't found anything on this list about
Combimatrix arrays, so I might just start by noting what I'm doing
with them, which may or may not be sensible. They are single color
arrays. It's not possible to estimate a spotwise background intensity
for these arrays, so I assume it is uniform over the whole array. I'm
using "normexp" for background correction, then quantile
normalization.

The arrays have some duplicated spots (sometimes 3x, sometimes 4x),
haphazardly arranged. I'm currently just averaging these duplicates.
How important is the business with ndups and duplicateCorrelation to
the final result, am I losing much by just averaging?

My second question is is it possible to obtain a "moderated" standard
error? If I do an MA plot, it's not clear why the top probes were
chosen, as they sometimes have smaller M than other probes. I'd like
to show error bars to give an indication of our confidence in each
spot. I imagine it's just a matter of multiplying or dividing a couple
of the vectors in the fitted model object, but I'm afraid that if I
try to guess which I will inadvertantly commit a small crime against
statistical sanity.

--
Paul Harrison

Victorian Bioinformatics Consortium / Monash University
pfh at logarithmic.net
jabber pfh at njhurst.com
icq 298231643
http://www.logarithmic.net/pfh/