[BioC] RE: Advice on print-tip normalization

Gene Cutler gcutler at amgen.com
Fri Sep 24 22:47:05 CEST 2004


Thanks for everyone's input.  I've been looking at the data more 
carefully and here's what I've found:

array geometry: grid = 12 rows x 4 cols, block = 28 rows x 27 cols

I see two periodicities.

1) When the intensities are ordered by block column, there are 27 
cycles of rising/falling intensities.  This gives 1344 spots per cycle 
which equals 48 columns of spots per cycle.  When these 27 cycles are 
overlayed and averaged together, I can see that each 1344 spot cycle 
contains another periodicity of 12 cycles: 112 spots per subcycle == 4 
columns per subcycle.

2) When the intensities are ordered by block row, there are 12 cycles 
of rising/falling intensities.  This gives 3024 spots per cycle which 
equals 112 rows of spots per cycle or 4 grid blocks per cycle.  When 
these 12 cycles are overlayed and averaged, I do not see any subcycles.

The coincidence of these cycles with whole row and column numbers 
suggests that it is a print tip effect.  I don't really understand 
print tip geometry yet, so the implications of the numbers above aren't 
clear to me.  Anyway, the marray boxplot shows what looks like a 
similar that goes away after printTipLoess normalization.  However, if 
I extract the normalized intensities and then plot those, the two 
cycles I describe above are still evident.






>>
>> Date: Fri, 17 Sep 2004 10:32:33 -0700
>> From: Gene Cutler <gcutler at amgen.com>
>> Subject: [BioC] Advice on print-tip normalization
>>
>> Hello.  I've just started using the marray package for processing a 
>> set
>> of spotted oligo arrays.  The arrays, when intensities or log ratios
>> are plotted against probe number, show a clear pattern of
>> rising/falling values with a periodicity equal to the grid block size
>> (~3600 spots).  I can see a similar periodicity in the printTip
>> boxplots generated with marray.  Running printTipLoess smoothes out 
>> the
>> boxplot nicely (and the MA plot also looks much nicer), but,
>> surprisingly, when I export the normalized values and plot them 
>> against
>> position, the grid block periodicity is little changed.
>>
>> I've tried different span values for the printTipLoess as well as
>> trying with or without scaling (e.g. printTipMAD), but nothing I do
>> seems to have much effect on this data artifact.
>>
>>




Gene Cutler
Research Investigator
Bioinformatics
Amgen SF



More information about the Bioconductor mailing list