search

courtiol / IsoriX

This is the GitHub repository dedicated to the development of the R package IsoriX
13 stars 6 forks source link

Add clustering on assignment? #147

Open courtiol opened 2 years ago

courtiol commented 2 years ago

Some packages do consider clustering (e.g. isoscat). It is based on Schoener's D metric which is a distance between pairs of assignment maps, followed by clustering approach.

While this is easy to implement, it would be good to implement a null model for the clustering to make sure the number of cluster obtained is not the sheer product of variation. Indeed, between-individual variation among samples coming from same location, once projected into an assignment map, may lead to select for more than one cluster, which would not be good.

The question is how to do that. Once option would be to simulate the situation assuming several samples with common origin (point location) picked at random across the landscape. We could consider several of such location to generate the distribution of the number of cluster under the null hypothesis (and thus p-value). That could be done in parallel to save some time as it would be costly. A minimal approach would be to consider the most likely location and only simulate number of cluster obtained in that particular situation (so no p-value, but expectation under H0).

courtiol commented 10 months ago

If you do think this issue matters to you, please add a thumb up!