[adegenet-forum] Find clusters on distance matrix or relative frequency.

[Jombart, Thibaut]

Hello, 

this is not a problem. You can run find.clusters on your composition profiles, just retain all PCs during the PCA step to ensure that all of your original information is preserved.

Alternatives would be using a single hierarchical clustering algorithm on the distance matrix (see the function hclust).

Best
Thibaut
________________________________________
From: adegenet-forum-bounces at lists.r-forge.r-project.org [adegenet-forum-bounces at lists.r-forge.r-project.org] on behalf of Francesco Montinaro [francesco.montinaro at gmail.com]
Sent: 19 July 2013 18:22
To: adegenet-forum at lists.r-forge.r-project.org
Subject: [adegenet-forum] Find clusters on distance matrix or relative  frequency.

Hi,
I have a matrix very similar to Structure/ADMIXTURE output but produced with a local ancestry algorithm.

For example for each population (10 populations) I have how much it "copies" from each of the other populations. Visually,It seems that there are 2 main clusters in composition of each populations, so I would like to cluster them.
Unfortunately my matrix is based on relative proportion, and I am afraid that find.cluster function is not suitable because it uses PCA. I was thinking to use the distance matrix of my original data, but I am not sure that it is mathematically correct.

In addition I would like to know the right number of cluster, so I cannot use kmeans, not natively at least.

Do you have any suggestion?

Thank you for your help.

Francesco