[BioC] On the validity of print-tip loess normalisation

[michael watson (IAH-C)]

Dear All

I have just seen a result which makes me either suspect the validity of
print-tip loess normalisation, or wonder if there is a bug in
bioconductor.

I am performing print-tip loess normalisation on a dataset which I have
already performed background correction on outside of bioconductor.
Here is my code:

> dataset[2896,1]@maRf
     chick1a_chick1.21406.bgc
[1,]                        6
> dataset[2896,1]@maRb
     chick1a_chick1.21406.bgc
[1,]                        0
> dataset[2896,1]@maGf
     chick1a_chick1.21406.bgc
[1,]                      228
> dataset[2896,1]@maGb
     chick1a_chick1.21406.bgc
[1,]                        0
> maM(dataset[2896,1])
     chick1a_chick1.21406.bgc
[1,]    
> ma <- maNorm(dataset, norm='printTipLoess')
            -5.247928
> ma[2896,1]@maM
     chick1a_chick1.21406.bgc
[1,]                 5.169925



As can be seen, my raw Cy5 value is 6 and the raw Cy3 value is 228.  My
raw ratio is 6/228 = 0.026, and my raw log2(ratio) is -5.2479.  All is
as expected.

But how on earth has the normalisation method turned my normalised ratio
to +5.169????  I clearly have a spot here which, if anything, is
down-regulated in the Cy5 channel.  Yet my normalised log ratio comes
out at +5 - which suggests a 32-fold increase in the Cy5 channel, from
data where cy5=6 and cy3=228.

I find this result pretty hard to accept....

Thanks in advance for the help

Mick