[BioC] filter on Human Gene U133 Plus 2

[Wolfgang Huber]

Il Nov/2/10 2:54 PM, wxu ha scritto:
>    Use MAS5, RMA, or other algorithms would be fine. The point is whether
> the gene filtering is proper. The raw p-values are not changed after
> filtering. The gene filtering only makes the FDR looking nicer.
> Different person may report a different FDR for the same gene list even
> with the same data set because of the different number filtered. If you
> want to get lower and nicer FDR, filter and reduce the total gene number
> to a small number. Do you think that is kind of cheating :)

Hi Wayne, Naima

in a nutshell, filtering is able to increase detection power, while 
maintaining type 1 error, if the filtering criterion is correlated with 
your test statistic (e.g. t-statistic) under the test's alternative, but 
independent from it under the null.

There are some important technical points on the application of this 
statement to different concrete test statistics, filtering criteria and 
type 1 error measures, which are discussed in the paper "Independent 
filtering increases detection power for high-throughput experiments", 
http://www.pnas.org/content/107/21/9546.long


	Best wishes
	Wolfgang


>
>
> Wayne
> --
>
> On 11/2/2010 4:27 AM, Naïma Oumouhou wrote:
>> Wayne,
>> Thank you for your response.
>> I thought it was a good idea to use the Detection MAS5 algorithm
>> because I remove genes not expressed or expressed a little.
>> Best regards
>>
>>   Le 29/10/2010 19:25, wxu a écrit :
>>> Naima,
>>> Why do you need to filter genes? For bad quality or some other
>>> reasons in your mind? Read my recent paper you may have an idea:
>>> http://www.ncbi.nlm.nih.gov/pubmed/20846437
>>>
>>> Wayne
>>> --
>>> Naïma Oumouhou wrote:
>>>> Dear Christian,
>>>>
>>>> I'm sorry to bother you again : I've got a question about filter on
>>>> Affymetrix Human gene U133 Plus 2 array.
>>>> I would like to find differentiallty expressed genes between 2
>>>> groups of patients (n1=7 and n2=6).
>>>> I have no experience in microarray analyses.
>>>> I read several publications and your xps vignettes but I don't know
>>>> what I have to do.
>>>> Some people filtered probesets using Detection MAS5 call:probesets
>>>> that aren't expressed in at least one sample using the  Detection
>>>> MAS5 algorithm are discarded.
>>>> What do you think about this filter?not tight enough?
>>>> After this filter and with my dataset,I still have 26 495
>>>> probesets?Is it too much?
>>>> Furthermore, in these remaining probesets, there are affymetrixx
>>>> control probesets. These probesets have to be removed?At which step?
>>>> After, I use the moderated t-statistics with BH correction. I find
>>>> no differentiallty expressed genes.
>>>> I wonder if I have to reduce the number of probesets with another
>>>> filter or with a "Detection filter" tighter?
>>>> Thanks for any help.
>>>> Naïma
>>>> Saisissez du texte, l'adresse d'un site Web ou importez un document
>>>> à traduire.<http://translate.google.fr/?tr=f&hl=fr>
>>>> <http://translate.google.fr/?tr=f&hl=fr>
>>>> Annuler<http://translate.google.fr/?tr=t&hl=fr>
>>>> <http://translate.google.fr/?tr=t&hl=fr>
>>>>
>>>>
>>>>      [[alternative HTML version deleted]]
>>>>
>>>>
>>>> ------------------------------------------------------------------------
>>>>
>>>>
>>>> _______________________________________________
>>>> Bioconductor mailing list
>>>> Bioconductor at stat.math.ethz.ch<mailto: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
>>>
>>>
>>>
>>>
>>
>
>
> 	[[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