Hi,
I am comparing different normalizations of Affymetrix microarrays:
1. expresso with rma, vsn, pmonly and medianpolish
2. calling rma and vsn with: eset=rma(Data); Set=vsn(eset);
I know that medianpolish does log2 transformation and vsn does log
transformation (base e), but when I compare the results, the
difference
between the two datasets is huge- not a matter of log-base.
Shouldn't the results be very similar?
Any help would be much appreciated.
Thanks,
Julia
On Fri, 6 Feb 2004, Julia Engelmann wrote:
>
> Hi,
> I am comparing different normalizations of Affymetrix microarrays:
> 1. expresso with rma, vsn, pmonly and medianpolish
> 2. calling rma and vsn with: eset=rma(Data); Set=vsn(eset);
> I know that medianpolish does log2 transformation and vsn does log
> transformation (base e), but when I compare the results, the
difference
> between the two datasets is huge- not a matter of log-base.
> Shouldn't the results be very similar?
>
Julia,
vsn doesn't use log (base) but a different trasnformation that
stabilizes
the variance, I have found the results to be quite different too,
they are meant to be especially if there is a strong dependency
between
variance and expression levels.
Best
Susan Holmes
Associate Professor
Statistics
Stanford
Hi Julia,
>Hi,
>I am comparing different normalizations of Affymetrix microarrays:
>1. expresso with rma, vsn, pmonly and medianpolish
>2. calling rma and vsn with: eset=rma(Data); Set=vsn(eset);
>
rma gives you expression levels on the log scale, whereas
vsn expects input values on the raw scale. So it doesn't make sense
to call vsn after rma like in 2. (apart from the question why you
should normalize your data twice - I guess in 1. you used rma only for
background correction; but calling rma like in 2. involves also
quantile normalization and medianpolish).
>I know that medianpolish does log2 transformation and vsn does log
>transformation (base e), but when I compare the results, the
difference
>between the two datasets is huge- not a matter of log-base.
>Shouldn't the results be very similar?
>
>Any help would be much appreciated.
>
>Thanks,
>Julia
>
>
Best,
Anja
On Sat, 7 Feb 2004, Anja von Heydebreck wrote:
> Hi Julia,
>
> >Hi,
> >I am comparing different normalizations of Affymetrix microarrays:
> >1. expresso with rma, vsn, pmonly and medianpolish
> >2. calling rma and vsn with: eset=rma(Data); Set=vsn(eset);
> >
> rma gives you expression levels on the log scale, whereas
> vsn expects input values on the raw scale. So it doesn't make sense
> to call vsn after rma like in 2. (apart from the question why you
> should normalize your data twice - I guess in 1. you used rma only
for
> background correction; but calling rma like in 2. involves also
> quantile normalization and medianpolish).
of cource you could do:
exprs(eset) <- 2^exprs(eset)
eset = vsn(eset)
exprs(eset) <- log2(exprs(eset))
this should make things look more alike.
>
> >I know that medianpolish does log2 transformation and vsn does log
> >transformation (base e), but when I compare the results, the
difference
> >between the two datasets is huge- not a matter of log-base.
> >Shouldn't the results be very similar?
> >
> >Any help would be much appreciated.
> >
> >Thanks,
> >Julia
> >
> >
> Best,
> Anja
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>
Rafael A. Irizarry wrote:
>On Sat, 7 Feb 2004, Anja von Heydebreck wrote:
>
>
>>Hi Julia,
>>
>>
>>>Hi,
>>>I am comparing different normalizations of Affymetrix microarrays:
>>>1. expresso with rma, vsn, pmonly and medianpolish
>>>2. calling rma and vsn with: eset=rma(Data); Set=vsn(eset);
>>>
>>>
>>>
>>rma gives you expression levels on the log scale, whereas
>>vsn expects input values on the raw scale. So it doesn't make sense
>>to call vsn after rma like in 2. (apart from the question why you
>>should normalize your data twice - I guess in 1. you used rma only
for
>>background correction; but calling rma like in 2. involves also
>>quantile normalization and medianpolish).
>>
>>
>
>of cource you could do:
>
>exprs(eset) <- 2^exprs(eset)
>eset = vsn(eset)
>
But the output of vsn is - at least approximately - on the natural log
scale.
>exprs(eset) <- log2(exprs(eset))
>
Therefore, to obtain log2-values, I would rather do
exprs(eset) <- log2(exp(exprs(eset))), or
exprs(eset) <- log2(exp(1))*exprs(eset)
Best,
Anja
--
Dr. Anja von Heydebreck
Max Planck Institute for Molecular Genetics
Dept. Computational Molecular Biology
Ihnestr. 73
14195 Berlin, Germany
heydebre@molgen.mpg.de phone: +49-30-8413-1168
http://www.molgen.mpg.de/~heydebre fax: +49-30-8413-1152
Dear all,
I am working in R a program for cross validation. I have a vector of
20
observations and I want to find all the possible combinations (20,4)
and
store them as data. I wonder if someone can help me.
Thanks in advance.
Makis
----------------------
E Motakis, Mathematics
E.Motakis@bristol.ac.uk
wrong place for this question, should go to r-help@r-project.org
anyway, i am not aware any R-function that gives all subsets of size
k. i
will use matlab to create it and input it back to R if you don't want
to
stick with matlab for the rest.
if you have some time to spare, you can create your own function by
implement the algorithms that generates all subsets of size k. a good
source for such algorithms is http://www.cs.sunysb.edu/~algorith/
good luck
Kenny
Kenny Ye
Assistant Professor
Department of Applied Math and Statistics
SUNY at Stony Brook
Stony Brook, New York 11794-3600
Phone (631)632-9344
Fax (631)632-8490
On Mon, 9 Feb 2004, E Motakis, Mathematics wrote:
> Dear all,
>
> I am working in R a program for cross validation. I have a vector of
20
> observations and I want to find all the possible combinations (20,4)
and
> store them as data. I wonder if someone can help me.
>
> Thanks in advance.
> Makis
>
> ----------------------
> E Motakis, Mathematics
> E.Motakis@bristol.ac.uk
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>
Hi E,
have a look at the function "nchoosek" from package vsn.
Best wishes
Wolfgang
E Motakis, Mathematics wrote:
> Dear all,
>
> I am working in R a program for cross validation. I have a vector of
20
> observations and I want to find all the possible combinations (20,4)
and
> store them as data. I wonder if someone can help me.
>
> Thanks in advance.
> Makis
>
> ----------------------
> E Motakis, Mathematics
> E.Motakis@bristol.ac.uk
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
--
-------------------------------------
Wolfgang Huber
Division of Molecular Genome Analysis
German Cancer Research Center
Heidelberg, Germany
Phone: +49 6221 424709
Fax: +49 6221 42524709
Http: www.dkfz.de/abt0840/whuber
