[adegenet-forum] interpretation sPCA

[Jombart, Thibaut]

Hello, 

you can't do this directly with k-means, but with DAPC. See vignette("adegenet-dapc"), last section, p.31. 

Cheers

Thibaut


________________________________________
From: Thomas, Evert (Bioversity-Colombia) [E.Thomas at CGIAR.ORG]
Sent: 14 October 2011 01:12
To: valeria montano; Jombart, Thibaut
Cc: adegenet-forum at r-forge.wu-wien.ac.at
Subject: RE: [adegenet-forum] interpretation sPCA

Dear Thibaut and Varleria,

Many thanks for your kind reply and the very useful explanations and suggestions! I have actually performed the ‘sliding windows’ approach but in a GIS environment with some interesting results… However, I have one new question: Is there any way to posteriorly determine to what cluster a plant individual that was not included in the k-means clustering is most associated?

Thanks for everything!

Cheers evert

From: valeria montano [mailto:mirainoshojo at gmail.com]
Sent: miércoles, 05 de octubre de 2011 06:00 a.m.
To: Jombart, Thibaut
Cc: Thomas, Evert (Bioversity-Colombia); adegenet-forum at r-forge.wu-wien.ac.at
Subject: Re: [adegenet-forum] interpretation sPCA

Hi again,

I see what you mean, I was superficially considering the fact that a clear spatial structure result with the spca could actually lead to spatially defined clusters in the dapc, but of course there is no warranty at all for that. I know your point about the (somehow) biological meaningless of the clusters, but still they are the genetic optimization of a specific dataset and I think this is, let's say, the "best achievable structure" in the contingency of someone's dataset. In very truth, in my view population structure is almost a philosophic concept (as much as fitness) and it is also true that any result in population genetic is usually the "dataset"'s point of view (with the genuine intuition of the one interpreting, of course). As for the summary statistics, a part from Fst, there are other useful ones, at least to get to know your data and also to support main findings. Btw the geoGraph package is really interesting...

Cheers

Valeria
On 5 October 2011 12:37, Jombart, Thibaut <t.jombart at imperial.ac.uk<mailto:t.jombart at imperial.ac.uk>> wrote:
Hello,

thanks for jumping into the discussion.

One has to be careful when playing with summary statistics derived from k-means. k-means finds groups which, by definition, maximise the Fst. So it is not clear how Fst values should be interpreted: real strong structuring, or indication or a good optimization procedure? In any case they cannot be tested, but that's not what's at stake here.

The problem using non-geographically constrained groups while looking for an origin is: how do you define the location of say, the associated Hs? Barycentre may do, but these groups may well not be geographically organised at all. One alternative though, following this idea, would be using spatially-constrained clustering.

True, moving windows are a bit arbitrary in that how the windows are defined is one's choice, but windows of say 3 different sizes could be defined and the results compared. Anyway I know of little spatial analyses which are not arbitrary (e.g. Neighbouring graphs in autocorrelation methods, data transformation in IBD/Mantel tests, etc.).

Cheers

Thibaut



________________________________________
From: valeria montano [mirainoshojo at gmail.com<mailto:mirainoshojo at gmail.com>]
Sent: 05 October 2011 01:25
To: Jombart, Thibaut
Cc: Thomas, Evert (Bioversity-Colombia); adegenet-forum at r-forge.wu-wien.ac.at<mailto:adegenet-forum at r-forge.wu-wien.ac.at>
Subject: Re: [adegenet-forum] interpretation sPCA

Hi there,

sorry for my usual gratuitous intervention. I just wanted to suggest the eventuality to use the dapc groups as an alternative to population labels (summary statistics of internal diversity on them would already help getting an idea of the overall situation). I actually think that at the intraspecific level previous grouping of individuals based on sample location criteria or whatever are usually quite biased. Genetic structure grouping might sound as a circular reasoning, but I have the feeling it is less arbitrary than any other approach. The "sliding windows" is definitely an alternative but still a bit arbitrary and maybe scheming, although it may be worth comparing the results of the both. Moreover, to individuate the most probable point of origin of the species, it may be useful to also explore the phylogeography as a support to the results obtained with the population approach.

Best

Valeria
On 4 October 2011 14:37, Jombart, Thibaut <t.jombart at imperial.ac.uk<mailto:t.jombart at imperial.ac.uk><mailto:t.jombart at imperial.ac.uk<mailto:t.jombart at imperial.ac.uk>>> wrote:
Dear Evert,

I don't think the existence of a cline can be used to infer the origin of an organism. Surely in this case the cline you obtain is compatible with a 'central' origin, but the origin could as well be at either extremities of the cline, or anywhere in between. All the pattern says is that gene flow is somehow negatively related to geographic distance. More generally, no multivariate analysis result is directional. It would be reassuring if the outcome of sPCA roughly match that of DAPC, although both methods are different. This can be easily checked by DAPC scores on the map. Discrepancies can be due to, for instance, the fact that non-spatial genetic structures are the strongest (then DAPC will pick that up first). Another one would be the absence of spatial structure. It is safer to perform a global.rtest (although it lacks power) and to check the screeplot of sPCA before interpreting structures.

Test the origin of your populations would need population-level data. The idea is that within-population diversity decreases when we get away from the origin due to repeated bottlenecks. If you don't have population data, one workaround would be using moving windows to map diversity geographically, and then use a simple optimisation procedure to find the 'optimal' origin. I don't know if this has been done before, so it might be newish. I have developed a package "geoGraph" (on Rforge, not on CRAN: https://r-forge.r-project.org/R/?group_id=348) which does this (apart from the moving windows) and has a vignette illustrating the whole process.

Cheers

Thibaut.



________________________________________
From: Thomas, Evert (Bioversity-Colombia) [E.Thomas at CGIAR.ORG<mailto:E.Thomas at CGIAR.ORG><mailto:E.Thomas at CGIAR.ORG<mailto:E.Thomas at CGIAR.ORG>>]
Sent: 03 October 2011 21:48
To: Jombart, Thibaut; Linda Rutledge; adegenet-forum at r-forge.wu-wien.ac.at<mailto:adegenet-forum at r-forge.wu-wien.ac.at><mailto:adegenet-forum at r-forge.wu-wien.ac.at<mailto:adegenet-forum at r-forge.wu-wien.ac.at>>
Subject: interpretation sPCA

Dear Thibaut,

I have a question regarding the interpretation of the sPCA scores as visualized in a color plot or interpolated lagged scores. I am working with intraspecific species data at continental level and found a strong gradient in my data  with a clear separation of a northern and southern group. Based on a number of grounds I believe that the center of origin of the species I am working with is located  at the "genotone" (or what to call this, I mean the grey area between both groups where the genetic differentiation is the steepest) . Does this make sense with the theory behind sPCA? I think the species moved north and south from the putative center of origin and developed into different genotypes which becomes apparent in the visualization of the sPCA...

And should the outcome of an sPCA be somewhat reflected in the outcomes of discriminant analysis of principal components or are these really two different methods? (I apologize for my ignorance)

Many thanks in advance

Evert
_______________________________________________
adegenet-forum mailing list
adegenet-forum at lists.r-forge.r-project.org<mailto:adegenet-forum at lists.r-forge.r-project.org><mailto:adegenet-forum at lists.r-forge.r-project.org<mailto:adegenet-forum at lists.r-forge.r-project.org>>
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/adegenet-forum