I'm getting similar results using DAPC, with a little informative bunch of mixed points. Instead, when I use PCA based in allele frequencies of populations (not individuals) the plot likely reflects my clustering results on STRUCTURE.

Well, the PCA is not very convincing of genetic structure, however you will have to plot the different axis to see if there is a clear differentiation of some clusters. When selecting the number of components to use, have a look on the explanation of each one, in your case the 6 retained components seems to give similar explanation (or at least the 4 first ones), so you will have to inspect each one with the same consideration (and not plotting only the two first ones) to maje conclusions.

I think there is not any real structure in my data. My settings were from the beginning more or less the same as René suggested, except of max K. I tried also the settings René suggested, including maxK=50, optim.a.score() which indicated 30 PCs to be retained, but without success. I think that the specific data do not support any genetic stratification and this is OK since pedunculate oak is anemogamous and the sampling area does not include any significant geographic barriers.

In addition, I just made a quick trial by including additional data from ca. 20 sessile oak stands and 6 downy oak stands to see if the method resolves at least the three species. Interestingly, K=2 was the most likely partition, which separated p. oak from the other two species (which are more related to each other). Setting K=3, the 3 species were not resolved! The two related species, downy and sessile oak stayed in the same group and the pedunculate oak was subdivided into two groups, but without any reasonable partition! I observed similar phenomena with STRUCTURE, which were due to asymmetry among group sizes of each unit (tested by simulating groups with different sizes). I described this in a recent paper. At the moment, I can't get deeper into DAPC, but maybe there are similar effects here, too, which seems interesting.

Hi there,

I've also dealt with samples that did not show any structure and find.clusers and the subsequent  DAPC nevertheless identified 4 different clusters. But as you said, division of individual into these clusters was rather arbitrary. Therefore, I did a DAPC on my actual samples straight away, rather than doing the find.clusters. In R,  I imported my data file (x) and then used the following:

dapc

Hi All thank you for wonderful discussion.

I would like to get suggestion from you all. How can I get member of population groups using adegnet package in R?

I used “adegnet package in R” and got five sub-population groups of the pathogen. Now I would like to get member isolates in each group. How can get the members of each group with membership probability? I have attached picture of cluster for your consideration. Thank you.

Hi Dipak,

to get the membership coefficients, you could use something like:

> round((mydapc$posterior),3)

This example gives rounded numbers with 3 decimals for each sample and each cluster of your dapc (here called 'mydapc').

I hope this helps !

Cheers,

Lise-Marie

Actually the structuring phase comes after to check the IBD (isolation by distance). For IBD the most commonly use criteria is Mental test but their is also some biases (Please read: Meirmans PG (2012) The trouble with isolation by distance. Mol Ecol 21:2839–284), so I would like to suggest first do IBD through RDA (Meirmans PG (2015) Seven common mistakes in population genetics and how to avoid them. Mol Ecol 24:3223–3231), if their is no IBD then go through the clustering, may be in your case you have  just one genetic group in actual not more than one.

@David Díez del Molino, you might be introducing that structure by calculating allele frequencies on populations decided a priori.