[BioC] Normalisation knocks my "housekeeping genes" off!

[michael watson (IAH-C)]

Hi

I have a small array that contains lots of spots for genes that *should*
show no differential expression, and guess what?  They don't!  The raw
data for these spots looks great, they all appear in a really tight
cluster on an M vs A plot and have a log2 ratio of just above zero,
which can easily be explained by dye bias.

HOWEVER, the general trend on the M vs A plot, when a loess line is
fitted, is not so good.  In fact, M shows a negative relationship with
A.  Therefore, when I loess normalise, it pushes all my log2 ratios for
the house-keeping genes up to around 1!  

I'm at a bit of a loss as to what to do - if I assume my house-keeping
genes should all be zero, I will push the trend of M decreasing with A
even further down.  If I correct for this trend using loess
normalisation, my house keeping genes will all display a two-fold
up-regulation.... :-(

Hmmmm.

Mick