"best" Ks and model components - question [fastSTRUCTURE]

[Sara Rocha]

Hi,

two use/interpretations questions here:

1) on trying to make sense of simple/logistic prior runs at a range of K's... I am aware of what is said in the paper: “For a new data set, we suggest executing fastSTRUCTURE for multiple values of K and estimating KE*  and K∅*C to obtain a reasonable range of values for the number of populations that would explain structure in the data, under the given model. To look for subtle structure in the data, we suggest executing fastSTRUCTURE with the logistic prior with values for values of K similar to those identified by using the simple prior.”

but would like to understand it a little better: 

what is the logic behind this? one should use this approach when the range in between Ke and Koc is large? and thus the logistic prior derived Ks may converge in a value? and because it is better at identifying shallow structure, they will provide in general better admixture plots within this range, is that it?

if in a range of K1-16, Ke/Koc with simple prior are always, say 4/5/6... any advantage at looking to admixture plots under logistic prior for those K's? (they can be slightly different...)

2) related, but not the same: I would also like to understand better how K's can be so different from the model components. In my case, with the logistic prior, I have often Ke as 16 (and I do realize that it tends to overestimate), but the nr of "components" in those admix plots is always 5/6 (which makes sense)... but I would understand it better if lk curves plateau at around 5/6 (like they do in the simple prior), but these have a trend of increasing towards K16 (plot bellow).. is this the normal behaviour? why this when most of the components are then 0.000002 across all individuals (in my real case). Is this the expected behaviour of this algorithm, or it may have any biological explanation?

many thanks in advance again!