[BioC] must one test all that is included in normalization?

[Naomi Altman]

Once the data are normalized, one uses only the genes that you are 
interested in - whether because the others are from a different 
species, or for any other reason that can be stated a priori without 
reference to the data at hand.

--Naomi

At 04:45 PM 6/6/2007, Maurice Melancon wrote:
>Hello,
>
>I worked with a single-colour multi-species array.  The dataset had limited
>options for transformation but disparately needed it, and vsn fit the bill
>nicely.  I was advised to include all non-empty spots in the normalization,
>even those that I am certain show hybridization due to non-target binding,
>because all spots contain useful information.  To qualify for a spot being
>'present' or 'absent', the creators of the array used the criteria of
>intensity above 2 SD of empty spots as their criteria.  This has proven to
>be far too liberal for my purposes, considering the nature of the array, and
>considering that while the dataset was in desparate need of scaling, the
>empty spots have little to no hybridization and thus the criteria for
>dropping spots prior to normalization or anything else is lax (imho).
>
>With vsn dataset in hand, the consensus seems to be among my collaborators
>to count only those spots that meet a certain criteria based upon the vsn
>data as present, and to test only those for differential expression.  One
>then calculates FDR based upon this subset of the data.  Is this fair?  Can
>you ignore spots used in normalization because you know them to be
>unreliable probes?  Or must this subset be eliminated prior to
>transformation?
>
>This is along a similar line to something posted earlier, where someone
>asked if it was allowable to normalize only a subset of the spots on an
>array, and to me the bigger question is how to fairly exclude spots that you
>are not interested in.
>
>With thanks,
>
>Maurice
>
>         [[alternative HTML version deleted]]
>
>_______________________________________________
>Bioconductor mailing list
>Bioconductor at stat.math.ethz.ch
>https://stat.ethz.ch/mailman/listinfo/bioconductor
>Search the archives: 
>http://news.gmane.org/gmane.science.biology.informatics.conductor

Naomi S. Altman                                814-865-3791 (voice)
Associate Professor
Dept. of Statistics                              814-863-7114 (fax)
Penn State University                         814-865-1348 (Statistics)
University Park, PA 16802-2111