[BioC] Quality checks / flagging

Yves Bastide Yves.Bastide at irisa.fr
Thu Sep 18 10:52:53 MEST 2003

Gordon Smyth wrote:
> At 05:04 PM 18/09/2003, Yves Bastide wrote:
>> Gordon Smyth wrote:
>>> See function read.maimages() and the section QualityWeights in the 
>>> limma package. You can read in any weight functions or flags that you 
>>> can calculate from the GenePix files, i.e.,
>>>   RG <- read.maimages(files,source="genepix",wtfun=myweightfun)
>>> where myweightfun is any function which you have defined which takes 
>>> Genepix data from one array as a matrix or data.frame and computes 
>>> from it a column of flags or quality weights. These weights then 
>>> become part of the component
>>>   RG$weights
>>> Note that read.maimages() produces an RGList object which is 
>>> essentially equivalent to a marrayRaw object. You can transform it to 
>>> an marrayRaw object if you want.
>>> Gordon
>> An even more elementary question (disclaimer: I am not a 
>> statistician): how to use these weights to make limma ignore some 
>> lines?  I naively tried with weights of 0 and 1, but it didn't work as 
>> expected.
> Weights of 0 and 1 do have the effect of ignoring the log-ratios for 
> spots with zero weight. I am not sure what you mean by "ignore some 
> lines". Please explain further what you did and what you expected to 
> happen which didn't.

Oh.  I zero-weighted flagged spots and spots for unwanted genes.  This 
resulted in about 600 genes to consider on each array, out of 4096. 
Then, the result of normalizeWithinArrays is strange, the final coef vs. 
lodds more so. (However, P-values and B are much better :) ).
Same for weighting-out only flagged genes, keeping 'bout 1/3 of them. As 
I'm just starting playing with statistic tools, I don't know what is 
expected and what is clearly wrong.

> Gordon
>> Regards,
>> yves

More information about the Bioconductor mailing list