[BioC] marray, weights and normalizations..

Henning Redestig redestig at mpimp-golm.mpg.de
Thu Apr 21 09:28:33 CEST 2005

> This is correct. Limma will do "loess" normalization for you but not print-tip-loess on such data.

I remedied this by "padding out" the Raw file with NA's which were 
systematically lacking spots in the last column of each block, this 
allowed me to do PT-loess in Limma which does not show the same strange 
behaviour as in marray when weights are used.

Another issue has occured though which I have seen on several datasets 
now related to using zero weights. Distributionally I get a whole lot 
more outliers leading to M values ranging between e.g. -200, 200 an 
effect I cant see when using weights of say, 0.1 instead (for all 
negatively GenePix flagged genes). Is this to be expected or am I doing 
something wrong?

 > As far as I know, maNormMain() only handles spot weights on a
 > single-array basis. I assume you are aware of that already.

Yes, I was aware of that since you have pointed that out on this list 
previously :)

Thanks for the reply!


Gordon Smyth wrote:
>> Date: Mon, 18 Apr 2005 12:13:03 +0200
>> From: Henning Redestig <redestig at mpimp-golm.mpg.de>
>> Subject: [BioC] marray, weights and normalizations..
>> To: bioconductor at stat.math.ethz.ch
>> Hi,
>> I am trying to use the Lapointe et al, PNAS 2004 data set from SMD
>> consisting of 112 arrays. These are not as I understand it LIMMA
>> compliant since the spots in the raw files are not directly in the
>> spotting order (some spots have been left out)
>>  and therefore I decided
>> to use the marray package which seem to be capable of handling even this
>> kind of formatting.
>> Using read.SMD() to import the data seems to work and image() can plot
>> the spots in spatial order indicating that the spotting order
>> information has been kept.
>> Problem arise when I try to normalize the data using maNormMain() as I
>> wish to weight the spots based on their flags.
> As far as I know, maNormMain() only handles spot weights on a 
> single-array basis. I assume you are aware of that already.
> Gordon
>>  Setting w to the weights
>> vector or NULL I get MA-plots as provided indicating a strong dependence
>> between A and M in the lower intensity range when weights are used
>> (lines are lowess fitted lines per print tip). Could anyone enlighten me
>> as to why this is the case? Isnt the whole point of the normalization to
>> remove any dependence between A and M?
>> The weights vector was set to 1 for flag=0, 0.1 for flag<=-50 and 0.01
>> for flag<=-75 (GenePix flagging conventions, and weights chosen 
>> arbitrarily)
>> Very thankful for help

More information about the Bioconductor mailing list