[BioC] limits/thresholds for arrayWeights in the limma package

[Matt Ritchie]

Hi Radek,

Arrays with weights less than 1 are less precise than average, while arrays
with weights greater than 1 are more precise (the sum to zero constraint in
the log-linear variance model enforces this).  So in your example, the array
which is assigned a weight of 4 is the most precise array in the experiment,
therefore I wouldn't recommend tossing it out.

The idea behind the array weighting method is to keep all of the arrays in
your analysis and weight the observations in the gene-wise linear models by
the estimated array weights which are intended to account for differences in
precision between different arrays.

I hope this explanation helps.  Best wishes,

Matt

>> Is there any limit ratio of between-array weight values, which one
>> can get using the arrayWeights() function in the limma package
>> <http://www.biomedcentral.com/1471-2105/7/261/abstract>? For
>> instance, if one array has the weight value - let's say - 4 and the
>> rest between 0.5 and 1.5, shall I keep the array and use it with
>> the weight array, or rather discard the one array and apply again
>> the arrayWeights() again and continue to results? I am particularly
>> interested in answers related to the mouse 430_2 chips.
> 
>> Regards, Radek
> 
>> Radek Blatny, MSc.
>> Institute of Molecular Genetics
>> Department of Mouse Molecular Genetics (Jiri Forejt unit)
>> Czech Academy of Sciences
>> Videnska 1083
>> 142 20, Prague
>> Czech Republic
>> Tel. (+420) 241 062 260
>> Fax (+420) 241 062 154
>> http://www.img.cas.cz/mmg
>> email: blatny at img.cas.cz
>> Skype name: blatny