[adegenet-forum] Inconsistent DAPC

[Jombart, Thibaut]

Yes, that's pretty much a recurrent problem: often there is no single clustering solution capturing the biological reality perfectly. See other posts about "true K" on the forum.

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 Vojtěch Zeisek [vojta at trapa.cz]
Sent: 09 July 2014 15:39
To: Adegenet R-Forum
Subject: Re: [adegenet-forum] Inconsistent DAPC

Hello, thank You very much.

Dne St 9. července 2014 10:49:04 jste napsal(a):
> Hello there,
>
> this is because you are using a fixed criteria to decide the group
> partitioning. This was only designed for simulated data analysis, or cases
> where one needs to analyse hundreds of datasets.
>
> Best for your to look at the BIC curves (over a couple of iterations) and
> assess which clustering solution you want to retain.

Ah, right. When running it like this
kfind <- find.clusters(genind, max.n.clust=75, stat="BIC",
choose.n.clust=TRUE, n.iter=100000000, n.start=10000, scale=FALSE,
truenames=TRUE)
table.value(table(pop(genind), kfind$grp), col.lab=paste("Inferred cluster",
1:length(kfind$size)), grid=TRUE)
I see there are more possible solutions... Interesting data set. :-)
Sincerely,
Vojtěch

> 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 Vojtěch
> Zeisek [vojta at trapa.cz] Sent: 09 July 2014 11:34
> To: Adegenet R-Forum
> Subject: [adegenet-forum] Inconsistent DAPC
>
> Hi,
> I was running DAPC on almost 500 individuals (over 60 populations) genotyped
> by 10 microsatellite primers as follows:
> kfind <- find.clusters(genind, n.pca=100, stat="BIC", choose.n.clust=FALSE,
> criterion="diffNgroup", max.n.clust=75, n.iter=100000000, n.start=10000,
> scale=FALSE, pca.select="nbEig", truenames=TRUE)
> table.value(table(pop(genind), kfind$grp), col.lab = paste("kfind",
> 1:length(kfind$size)))
> dapc <- dapc(genind, kfind$grp, n.pca=35, n.da=20, center=TRUE, scale=FALSE,
> var.contrib=TRUE, pca.info=TRUE, pca.select="nbEig", truenames=TRUE)
> scatter(dapc, main="DAPC", posi.da=“bottomleft“)
> When I did it for first time, I've got nice partitioning to 3 clusters,
> quiet similar to output of Structure and PCoA. When I tried it second time
> with same parameters, I've got division to 12 groups. So I wonder how is it
> possible and how to trace to make it reproducible and find out why it
> shifted so much. Sincerely,
> Vojtěch
--
Vojtěch Zeisek
http://trapa.cz/en/

Department of Botany, Faculty of Science
Charles University in Prague
Benátská 2, Prague, 12801, CZ
http://botany.natur.cuni.cz/en/

Institute of Botany, Academy of Science
Zámek 1, Průhonice, 25243, CZ
http://www.ibot.cas.cz/en/

Czech Republic