[adegenet-forum] Connection network choice & error messages

Jombart, Thibaut t.jombart at imperial.ac.uk
Wed Jun 24 22:59:44 CEST 2009


Dear Sofie,

thanks for your post.

> Dear all,
>
> This is my first time working with connection networks, so please be indulgent...
> Some questions:
> 1) Can anyone recommend some literature discussing the properties, (dis)advantages etc of the different connection networks available in sPCA?

Well, most of the graphs available via chooseCN are described in this book:
Legendre, P. & Legendre, L. Numerical ecology Elsevier Science B. V., Amsterdam, 1998

This may be a start.

> 2) it was recommended that I use the inverse distance connection network, which then requests several parameters (exponent and min distance), any suggestions/rules on how to decide on these (by the way, I entered my data as latitudes and longitudes)?

The exponent is there for special cases where connectivity would be expected to decrease, say, like the square of the Euclidean distance between objects. This was probably overkill, since I did not meet such case up to now. I think it is safe to leave it to 1. The minimum distance is used to set the maximum proximity between two objects, which is 1/dmin. Otherwise, you might have cases where the distance is null, and you would end up with useless +Inf... For instance, you can use 1 to scale your neighbouring weights to 1.

Concerning lat/lon, the spca manpage (?spca) describes the procedure to follow (argument xy). 

> 3) what criteria should be used to decide on the number of axes selected in the spca object?

The sPCA paper tackles this point a bit, but you may essentially want to have a look at the sPCA tutorial, where the choice of interpreted structures is discussed and illustrated.  In particular, use the barplot of eigenvalues, the decomposition of eigenvalues (screeplot.spca), and global/local tests (global.rtest, local.rtest).

Best,

Thibaut.

> Thanks for any constructive input!
>
> Sincerely,
> Sofie
>
> Sofie Vandewoestijne
> Postdoctoral fellow
> Biodiversity Research Centre
> Catholic University Louvain
> Place Croix du Sud 5
> B-1348 Louvain-la-Neuve
> Belgium
>
> Tel: 32 (0)10 473454
> Fax: 32 (0)10 473490
>
> P Please consider your environmental responsibility before printing this e-mail



More information about the adegenet-forum mailing list