[BioC] CGH segmentation algorithms (snapCGH package)

[Ramon Diaz-Uriarte]

Dear Joao,

I am not sure your idea would get you what you want. The issue you rise has 
been addressed before in other fields where you want to combine several 
disgnostics or experts or classifiers. For example, if you search 
for "mixtures of experts" you'll find a lot of hits. Machine learning 
textbooks often have one or more chapters devoted to this issue. You might 
try to derive a (meta-) algorithm that combines the output from those three, 
based upon some training data (of a structure similar to yours) for which the 
truth is known. In some cases, using algorithms which return a probability 
makes combination simpler (i.e., if each algorithm tells you how sure it is 
of its assessment); that is not really the case with any of the three you 
mention.

Best,

R.





On Monday 19 November 2007 10:46, João Fadista wrote:
> Dear all,
>
> I used 3 different segmentation algorithms (DNAcopy, GLAD and HomHMM from
> snapCGH package) to assess copy number variation in my data and I want to
> have a criteria to choose the best CNVs found. There are cases where all
> the 3 algorithms find the same CNV, cases where only 2 algorithms find the
> same CNV and cases where only 1 algorithm finds a CNV.
>
> Somebody could argue that a good criteria should be to exclude CNVs found
> with only 1 algorithm but I found that sometimes those CNVs are really
> there when I make a visual inspection of the data. I thought about making a
> Mann-Whitney test for each CNV to compare the log2-ratios between n probes
> from the CNV with (c-n) probes from the control region, where 'c'
> represents the total number of probes for that chromosome that are not in
> any CNV on that array. After correcting for multiple testing, a p-value is
> assign for each CNV giving me a criteria to choose the best CNVs found.
>
> Any comments on my approach and alternative ones will be very much
> appreciated. Thanks in advance.
>
>
> Best regards,
> João Fadista
>
> 	[[alternative HTML version deleted]]

-- 
Ramón Díaz-Uriarte
Statistical Computing Team
Centro Nacional de Investigaciones Oncológicas (CNIO)
(Spanish National Cancer Center)
Melchor Fernández Almagro, 3
28029 Madrid (Spain)
Fax: +-34-91-224-6972
Phone: +-34-91-224-6900

http://ligarto.org/rdiaz
PGP KeyID: 0xE89B3462
(http://ligarto.org/rdiaz/0xE89B3462.asc)



**NOTA DE CONFIDENCIALIDAD** Este correo electrónico, y ...{{dropped:3}}