Admixture analyses

With Cory's help I finally got my data into the correct format to get it into Admixture (took several hours of hitting my head against the wall). I ran Admixture on two different datasets. Both datasets involved using only SNPs that had been filtered for Good contigs, over 2.5kb, and filtered with maf=0.15 and for LD. In the first dataset I included all individuals we have so far (minus the 6 I dropped for low coverage and also not including the individual from D Barb). The second dataset included only one individual per super clone/pond/season (super clones subset). According to the Admixture manual the way to determine your best K is to look at CV error, and choose the K where that value is lowest. First I ran the Admixture analysis using Ks of 1-12 on the full dataset. Here is the distribution of CV error as well as log likelihood.

For both of these you can see that there is a change in the curve at K=3. However, the CV error continued to get lower and never went back up again. This is not how it is supposed to happen (according to the manual).
Here are some output graphs from Admixture on the full dataset. Sorted by pond and year.

K=2

K=3

K=4

K=5

K=6

K=7

Ok. So I was thinking the problem with the CV error analysis for determining the best K is probably due in part, in not all, to the high clonality. So, I reran everything with subsetting super clones. Here are the CV error and Log Likelihood distributions.

So from this, we again see an inflection at K=3. The lowest CV error is at K=5, but it is not that different from K=3 (perhaps?). But it seems the best K is probably somewhere between K=3 and K=5. Here are the outputs with the super clones subset.

K=2

K=3

K=4

K=5

K=6

K=7

So, what do we take from all of this? First, D8 is never a single population. The two "clades" in D8 always appear to fall out in separate populations. If you look at K=3 then DBunk (also D Oily, D Cat, R Dramps to some extent) does appear to be generally admixed between the 2 populations that are seen in D8, and D10 is divergent. In general, DBunk always appeared admixed, with few if any solid bars, regardless of the value of K. Again, this also appears to hold true for D Ramps and D Oily, perhaps a bit less so for D Cat. So maybe D Bunk are admixed between the two super clones? If so, it appears there are a few D8 unique clones that may also be admixed (D8_Spring_2017_134, D8_Spring_2017_214, D8_Spring_2016_8.25, and D8_Spring_2016_8.26). Still not sure how much I believe these analyses. They look an awful lot like the PCA, which I guess is what we would expect, but somehow I was hoping for something more?