[BioC] edgeR MDS

[Gordon K Smyth]

Dear Jose,

The plotMDS() limma function doesn't compute classical Euclidean 
distances, not by default anyway.

Cheers
Gordon

---------------------------------------------
Professor Gordon K Smyth,
Bioinformatics Division,
Walter and Eliza Hall Institute of Medical Research,
1G Royal Parade, Parkville, Vic 3052, Australia.
http://www.statsci.org/smyth

On Thu, 23 May 2013, Jose M Garcia Manteiga wrote:

> Dear Gordon,

> I was always wondering about the same question since I am not a 
> statistician. What I found reading on the internet is that MDS uses 
> distances to calculate eigenvectors, so you can do MDS although you do 
> not know the absolute coordinates but using the relative distances. But 
> if you have the original values, then standard Principal Component 
> Analysis and classical (metric) MDS should give the same components. But 
> please correct me if I am wrong. That would explain why plotMDS and PCA 
> score plots are that often that similar when used with limma.

> Thanks for the posts.
> Jose
>
>
>
> On May 23, 2013, at 1:53 AM, Gordon K Smyth <smyth at wehi.edu.au> wrote:
>
>> Dear Manoj,
>>
>> plotMDS does not do PCA.  As the documentation says
>>
>> "This function is a variation on the usual multdimensional scaling (or
>> principle coordinate) plot".
>>
>> Statisticians are sometimes not very imaginative when choosing names for
>> things, but PCA is an abbreviation for "principle component analysis"
>> which is not the same as "principle coordinate analysis".
>>
>> Best wishes
>> Gordon
>>
>>> Date: Tue, 21 May 2013 11:55:36 -0700 (PDT)
>>> From: Manoj Hariharan <h_manoj at yahoo.com>
>>> To: "Bioconductor at r-project.org" <Bioconductor at r-project.org>
>>> Subject: [BioC] edgeR MDS
>>>
>>> Hello,
>>>
>>> I am working on edgeR version 3.2.3.
>>>
>>>> From the documentation, I guess the "plotMDS.DGEList" is similar to
>>>> PCA. The manual mentions that "Distances on the plot represent
>>>> coefficient of variation of expression between samples".
>>>
>>> Is it possible to get a value of proportion of variance explained from
>>> each dimension (component)? Also, is it possible to use the DGEList to
>>> make a 3D PCA plot?
>>>
>>> Thanks,
>>>
>>> Manoj.
>>>
>>>
>>> Manoj Hariharan
>>> Staff Researcher
>>> The Salk Institute for Biological Studies
>>> La Jolla, CA 92037
>>> Office: 858.453.4100 x2143

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}