[adegenet-forum] testing sPCA differences

Jombart, Thibaut t.jombart at imperial.ac.uk
Mon Jul 14 11:42:16 CEST 2014

Hi Xavier, 

yes, it makes sense. I think Stéphane Dray (Cced) did something along these lines a while back. The basic idea is to use a spatial grid to define new spatial units which can be compared across different categories of individuals. 

Building up a test for this should not be too complicated, if we're talking about one single variable. You can just use the (possibly standardized) difference in Moran'I values, and use permutations to generate the reference distribution. Type I error will (by construction) be OK, you'll only have to assess the power of the test.

When considering more than one variable, things get complicated as basic tests are already becoming more scarce. As for sPCA PCs, the problem is that they are already optimized, so most simple tests are circular. The non-independence of the tests of different PCs also is a problem. Valeria Montano started working on such issues and may be interested in this discussion. Last time I checked she was on this ML but I'll Cc her to bully her into the discussion.


From: Xavier Giroux-Bougard [x.giroux.bougard at gmail.com]
Sent: 11 July 2014 17:40
To: Jombart, Thibaut
Subject: Re: [adegenet-forum] testing sPCA differences

Thanks for the answer Thibault,

I too had thought of varying spatial locations as a problem. However, there is a way to circumvent this problem if we move away from an individual based approach, which of course requires a species with more discrete groupings. In this case, if we build two 'genpop' objects using individuals seperated from their initial population based on some other characteristic (ie sex, age, morph, etc...), then the connection network remains identical if we use the population locations. While the spatial weights remain constant, we are left with two distinct allelic frequency tables. Admittedly we are losing power by moving away from and individual based network, but then we could stand to gain a useful comparative method.

Let me know if you think that makes sense, and if you think there would be an easy way to implement this.

Merci pour votre aide, et votre temps!

Xavier Giroux-Bougard

PS If the concept holds, I would be glad to share my coding explorations (as amateur as they might be!)

On Fri, Jul 11, 2014 at 5:14 AM, Jombart, Thibaut <t.jombart at imperial.ac.uk<mailto:t.jombart at imperial.ac.uk>> wrote:

Hi Xavier,

such a test will not be possible using approaches directly relating to the sPCA. This is because the range and null value of Moran's I (part of the optimized criteria, measuring the spatial structures) depend on the spatial weights. Therefore, Moran's I based approaches cannot compare spatial structures of different groups of locations/individuals.

From: adegenet-forum-bounces at lists.r-forge.r-project.org<mailto:adegenet-forum-bounces at lists.r-forge.r-project.org> [adegenet-forum-bounces at lists.r-forge.r-project.org<mailto:adegenet-forum-bounces at lists.r-forge.r-project.org>] on behalf of Xavier Giroux-Bougard [x.giroux.bougard at gmail.com<mailto:x.giroux.bougard at gmail.com>]
Sent: 10 July 2014 22:14
To: adegenet-forum at lists.r-forge.r-project.org<mailto:adegenet-forum at lists.r-forge.r-project.org>
Subject: [adegenet-forum] testing sPCA differences

Dear adegenet forum,

does anybody know if there is a permutation test implemented in adegenet (or other R packages) that could differentiate the sPCA solutions of two (or more) subsets of a genetic dataset, for example to compare genetic structures across sexes or ages, etc... For example, Smouse et al. (2008) developed a heterogeneity test to differentiate two correlograms.

I know there are the global and local statistics, but is there something similar that could help differentiate the spatial genetic structure between two groups in sPCA?



More information about the adegenet-forum mailing list