[BioC] nsFilter

[Vincent Carey]

On Mon, Nov 22, 2010 at 3:57 AM, Peevi Ijkl <ipeevi at yahoo.com> wrote:
> Hello List
> I wanted to perform nonspecific filtering on my expression set. i gave the
> following code
>>f1<-function(x) IQR > 0.5)

This question is not very clear, and the coding is peculiar and
nonsyntactic -- there is an unmatched parenthesis in the line defining
f1.  As written above "IQR" names an unbound symbol in function f1.
in a fresh standard R session, "IQR" is bound to a function in the
stats namespace, so unless there is some globally defined vector with
this name, the above code would compare a function to 0.5, probably
unintended.

If you want to use filtering on the actual values of the gene-specific
IQR, the approach that is most consistent with the genefilter examples
is

IQRlb = function(lb=0.5, na.rm=TRUE) function(x) IQR(x, na.rm=na.rm)>=lb
     # note that IQRlb is a closure, a function returning a function
IQRfiltP5 = filterfun(IQRlb(.5))  # allows flexible setting of bound and
                                                 # introduction of
additional tests
set.seed(1234)
mm = matrix(rnorm(100,0,.45),nr=10)
kp = genefilter(mm, IQRfiltP5)
table(kp)

> table(kp)
kp
FALSE  TRUE
    6     4

This is NOT consistent with the default behavior of nsFilter, in which
the distribution of the gene-specific dispersion measure is computed
for all genes that are relevant in the context of the parameters of
the nsFilter call, and genes are kept if their dispersion exceeds a
specified quantile (by default the median) of the distribution of the
dispersion measure.  This is owing to the default setting of
"filterByQuantile" in nsFilter.

to emulate this aspect of nsFilter in a very simple way, you could do
the following

medIQR = median(apply(mm,1,IQR))
kp = apply(mm,1,function(x)IQR(x)>=medIQR)

but note that nsFilter accepts and returns an ExpressionSet as
representation of the data.
Furthermore, computation of IQR in nsFilter is performed by mechanisms
other than stats::IQR.  The numerical definition of quantiles can be
intricate.



>>ff<-filterfun(f1)
>>selected<-genefilter(data2,ff)
>>sum(selected)
>  i knw this is for genefilter..but can the above be used for non specific
> filtering too?
>
> please help
>
> peevi
>
>
>
>
>        [[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
>