[BioC] Normalization

[Naomi Altman]

Dear Caryn,

I have not evaluated the effect of normalization when there is a 
blocking factor, and I am not aware of any reference about this.
I guess you could normalize within blocks if you have a randomized 
complete block design and use block as a factor.  I would not want to 
do this with an incomplete block design.

I guess as a statistician the issue was obvious to me:

Analysis of differential expression compares the variance between 
treatments to the variance within treatments.
Since the effect of normalization is to reduce the variance among the 
arrays normalized together, and since separate normalizations are likely to
introduce variance between the groups, it seems obvious that the net 
effect of normalizing within treatment group will be to increase the 
false detections.

I guess I do not have a good feel for what needs to be put into the 
literature.  A number of issues that I thought were obvious ended up 
being worth the effort of experimental follow-up and a paper.  People 
seem to feel that microarray data will behave differently than the 
data scientists and statisticians have worked with for the past 
century, but I have seen no evidence that this is so.

--Naomi

At 04:01 PM 3/1/2006, Caryn M Thompson wrote:
>Naomi,
>
>       A copy of your recent posting to the Bioconductor discussion list
>was forwarded to me.  I haven't seen much in the literature re:the issue
>of normalizing within conditions versus across all arrays - could you
>point me to some good references?  Also, have you done anything to
>evaluate the effect of various normalization schemes when a blocking
>factor is involved?  I've seen a recent argument to suggest that
>normalization within blocks is appropriate provided a blocking factor is
>included in the model (assuming a linear models approach is taken for
>analysis).
>
>Best regards,
>Caryn Thompson
>
>
>
>Caryn M. Thompson, PhD
>Associate Professor
>Director of the Statistical Consulting Center
>Department of Bioinformatics and Biostatistics
>School of Public Health and Information Sciences
>University of Louisville
>555 S Floyd St.
>Louisville, KY  40292
>Phone: (502) 852-1888
>Fax: (502) 852-3294

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