[BioC] different normalized results using oneChannelGUI, and oligo packages

[Henrik Bengtsson]

A pairwise plot reveals difference too:

x <- exprs(OligoEset);
y <- exprs(rma-sketch);

x <- as.vector(x);
y <- as.vector(y);
plot(x,y, pch='.");
abline(a=0, b=1);

/Henrik

2009/11/16 Javier Pérez Florido <jpflorido at gmail.com>:
> Dear Benilton and Raffaele,
> I've tried what you suggested and I got the following results:
>
>    * Comparison between rma and rma-sketch using APT tools:
>          o apt-probeset-summarize -a rma -p Hugene-1_0-st-v1.r3.pgf -c
>            Hugene-1_0-st-v1.r3.clf -m Hugene-1_0-st-v1.r3.mps -o rma
>            --cel-files cel_list.txt
>          o apt-probeset-summarize -a rma-sketch -p
>            Hugene-1_0-st-v1.r3.pgf -c Hugene-1_0-st-v1.r3.clf -m
>            Hugene-1_0-st-v1.r3.mps -o rma-sketch --cel-files cel_list.txt
>
>    The error given by abs(exprs(rma)-exprs(rma-sketch)) / num_points is
>    1.4147e-004
>
>    * Comparison between rma using APT tools and rma using oligo:
>          o apt-probeset-summarize -a rma -p Hugene-1_0-st-v1.r3.pgf -c
>            Hugene-1_0-st-v1.r3.clf -m Hugene-1_0-st-v1.r3.mps -o rma
>            --cel-files cel_list.txt
>          o OligoEset<-rma(OligoRaw,target="core")
>
>    The error given by abs(exprs(rma)-exprs(OligoEset)) / num_points is
>    1.8915
>
>    * Comparison between rma-sketch using APT tools and rma using oligo:
>          o apt-probeset-summarize -a rma-sketch -p
>            Hugene-1_0-st-v1.r3.pgf -c Hugene-1_0-st-v1.r3.clf -m
>            Hugene-1_0-st-v1.r3.mps -o rma --cel-files cel_list.txt
>          o OligoEset<-rma(OligoRaw,target="core")
>
>    The mean error given by abs(exprs(rma-sketch)-exprs(OligoEset)) /
>    num_points is 1.8915
>
> Surprisingly, the difference between APT tools and oligo are quite high
> (you can notice it when having a look at the normalized data).
> Is there anything wrong above?
> Thanks,
> Javier
>
>
>
> Benilton Carvalho escribió:
>> Thanks for the heads up, Raffaele.
>>
>> My suggestion was Javier to try "-a rma" manually, instead of "-a
>> rma-sketch" (which appears to be what oneChannelGUI uses - at least
>> that's what I saw when skimmed the source ).
>>
>> cheers,
>>
>> b
>>
>> On Nov 12, 2009, at 6:32 AM, rcaloger wrote:
>>
>>> Hi,
>>> dear Benilton,
>>> you are right oneChannelGUI uses sketch-quantile. APT-tools are used
>>> by oneChannelGUI to calculate intensities.
>>> Actually oneChannelGUI uses a graphical interface to execute
>>> apt-probeset-summarize -a rma -p PGFFILE.pgf -c CLFFILE.clf  -m
>>> MPSFILE.mps -out rmasketch *.cel
>>> cheers
>>> Raffaele
>>>
>>>
>>>> Javier,
>>>
>>>> my knowledge on the oneChannelGUI is pretty limited, but it seems to
>>>> me that it does either rma-sketch or iter-plier.
>>>
>>>> oligo will do the "regular" rma, ie. no sketch normalization. Now, if
>>>> you could download APT and run it yourself (I'm not sure how it is
>>>> packaged in oneChannel), you could try the following:
>>>
>>>> apt-probeset-summarize -a rma -p PGFFILE.pgf -c CLFFILE.clf  -m
>>> MPSFILE.mps -out rmasketch *.cel
>>>
>>>> And you may get more similar results... I'm not saying they'll be
>>>> identical, but when I implemented this for oligo the mean relative
>>>> difference between oligo and APT was something like 0.001.
>>>
>>> b
>>>
>>> On Nov 10, 2009, at 3:30 PM, Javier P?rez Florido wrote:
>>>
>>>
>>>>> Dear all,
>>>>> I have normalized a GeneChip Human Gene ST 1.0 dataset using oligo
>>>>> package and oneChannelGUI. I've checked that the log2 normalized
>>>>> values
>>>>> are different (not so different, but different). When applying limma
>>>>> on
>>>>> them, the statistic values (p-value, M, B and statistic t) are also
>>>>> different.
>>>>>
>>>>> I've read that oneChannelGUI uses APT tools to summarize data and,
>>>>> maybe, that is the point: they produce similar but different
>>>>> normalized
>>>>> values that have an effect in the final list of differentially
>>>>> expressed
>>>>> genes.
>>>>>
>>>>> For oligo, I've normalized using rma method and core as a target:
>>>>> OligoEset<-rma(OligoRaw,target="core")
>>>>> For oneChannelGUI, I've used the option "Load gene 1.0 ST arrays from
>>>>> .CEL files."
>>>>>
>>>>> In both cases, the limma parameters are the same.
>>>>> So, I suspect the APT tools used by oneChannelGUI has influence on the
>>>>> data. Is there a way to normalize the raw data in OneChannelGUI to
>>>>> select the subgroup (core) instead of using APT tools? I think this is
>>>>> only available in Exon Arrays.
>>>>>
>>>>> Or may be I am wrong with these conclusions.
>>>>> Any tips?
>>>>> Thanks,
>>>>> Javier
>>>>>
>>>>
>>> --
>>>
>>> ----------------------------------------
>>> Prof. Raffaele A. Calogero
>>> Bioinformatics and Genomics Unit
>>> Dipartimento di Scienze Cliniche e Biologiche
>>> c/o Az. Ospedaliera S. Luigi
>>> Regione Gonzole 10, Orbassano
>>> 10043 Torino
>>> tel.   ++39 0116705417
>>> Lab.   ++39 0116705408
>>> Fax    ++39 0119038639
>>> Mobile ++39 3333827080
>>> email: raffaele.calogero at unito.it
>>>       raffaele[dot]calogero[at]gmail[dot]com
>>> www:   http://www.bioinformatica.unito.it
>>> Info: http://publicationslist.org/raffaele.calogero
>>>
>>
>>
>
>
>        [[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
>