Estimate the best K

Question

Estimate the best K

gtonkinhill opened this issue 9 years ago · 6 comments

A function that runs the method of Evanno and plots ln(Pr(X|K) values in order to identify the k for which Pr(K=k) is highest (as described in STRUCTURE's manual, section 5.1)

Answer 1 · 2016-03-15T05:25:56.000Z

I also think an interesting possibility would be to pool results from various tools here. Not exactly sure how to go about this at this stage.

Answer 2 · 2016-03-15T23:00:47.000Z

clumpak does something like that, although I haven't looked at it yet

Answer 3 · 2016-03-30T03:48:41.000Z

for admixture is this the change point in the CV graph?

Answer 4 · 2016-05-06T01:22:28.000Z

I think we can close this now, unless we want to put something in for Admixture?

Answer 5 · 2016-05-06T01:25:42.000Z

looks good. i'll add something in for admixture before closing that uses admixtures cross-validation method to select the min.

Answer 6 · 2016-05-10T01:25:25.000Z

I've added in admixture CV graph. Also throws error if the user hasn't added a log file.

With the repeated runs for STRUCTURE would it be possible to do something akin to https://en.wikipedia.org/wiki/Wilcoxon_signed-rank_test for model selection?

Also I don't know how mathematically sound this would be... but if you're doing repeated structure runs kinda akin to doing a 'bootstrap' like procedure ---> could do likelihood ratio tests?