I would try to analyze the same data with BAPS for instance and see whether the results merge in K=16 or not.

With the data you have supporter I would say that the evidence of a population structure of K=16 seems pretty likely. Can you upload the figure of the 'plot in a single line' instead in 'multiple lines'?

They are plotted in a single line. This diagram just shows q-values populationwise i.e. each box shows the membership proportion of each population. I made it using Distruct (Rosenberg 2004).

Hello,

Did you use structure harvester to calculate the Delta K values? The output looks like you did, but if not you should try using it and see if you get the same approximations.

Based on the banding pattern that you recovered I think that structure is having a issue distinguishing between any of the groups and your result is a K of 1. I agree you should try BAPS and see if you have a similar problem.

Did you compute pairwise FST values? Latch et al. (2006) Relative performance of Bayesian clustering software for inferring population substructure and individual assignment at low levels of population differentiation in Conservation Genetics indicated that STRUCTURE could not distinguish and was somewhat inconsistent below an FST of 0.03 . If your values are below that it may not be informative to run STRUCTURE at all.

Sorry, I meant the plot of the q-values of every individual separated by populations. I think that is lot more visual than the average values by populations.

Hi Timothy and David,

Yes, I used Str. Harvester for inferring ΔΚ values. David, here is an example of a K16 run (not averaged using CLUMPP and DISTRUCT because this would require too much time). As you see, structure is not that obvious. Or maybe with some imagination :)

Multilocus pairwise Fsts are in the magnitude of 0.01 to 0.03; indeed very low, but not unexpected for populations of an anemogamous tree growing not far from each other (max 200 km).

And yes, I had tried BAPS with slightly different data. Because I have 3 oak species, I had to define purebreds of each species first before I started analyses within species. BAPS showed most likely partition for K=1, even when coordinates were included. Now I slightly changed my criteria of choosing purebreds (I made them a bit more stringent, based on simulation based runs; btw recently published). I assume that it will do the same now. Depending on time, I will maybe give it a last try and tell you what happened.

Cheers,

Charalambos

Ok, that structure plot doesn't point to K=16.

:)

OK, I think there is not strong enough evidence about structure.

I just tried some test runs with spatial clustering of individuals with BAPS setting maximum K=1, 2, 5, 10, 20 and it always finds optimal partition for K=1.

I posted the question also in the Structure discussion forum. Here is the link: https://groups.google.com/forum/#!topic/structure-software/HhWud1qnOUY. Maybe there are some interesting aspects for you there, as well.

Regarding pairwise Fst, I'm aware about the limitations caused by sample size.

Thank you for your response.

Charalambos

The Evanno method will always behave strangely when your K = 1 using the normal ln method (as in, when you have no obvious structure). Obviously the Evanno method cannot give an answer of K = 1. Your results don't indicate structure, which is interesting in itself. I would also go with Kevin's advice and see how your data looks using PCA/DAPC in adegenet in R, could be informative.

Thanks a lot, Allan and Kevin, for your hint for PCA/DAPC. I will give it a try and inform you when I have results. I hope it won't take too long to get familiarized with adegenet... :)

Yes, I tried DAPC (without spending too much time), "found" some evidence about structure, but I'm still not convinced. I posted a separate question.

Cheers,

Charalambos

what about using Structurama and comparing the clustering of it to Structure?

Guys, i have seen a data wherein there is no peak in the L(K) graph [K =1-10]. what does it actually mean? the PCA shows segregation of samples based on geographic distribution. the experiment had 70 samples grouped into 18 population from very distant locations. can anybody explain me, what might be the reason for absence of peak?

Have you used Evanno's method, delta K? Use Structure Harvester for this. If there's no peaks perhaps you markers are not strong enough to show population structure. How many markers are you using? Have you tried playing around with the program or are you just using default settings?

I haven't tried the delta K method yet. As far as now, I have just used the default settings. And you know what, these are dominant marker data not SSR

Are you using AFLPs? I've only used the program with microsatellites so i'm not too sure if other markers will act differently. I would certainly try other settings and delta K. See what that does. Have you tried other programs such as BAPS? I'm sure you can use AFLPs on that program as well and plenty of people would say it's a better program. Good luck.

Thank you Mr. Sam. i will check what you have mentioned