fastStructure - strange results when choosing best K

[Alberto Fameli]

Disclaimer: A similar question was posted by Yifang Tan in May 2020 but received no replies, so here I go...

Hi all, 

I have an big dataset (1700+ samples and 64,000+ SNPs) and I ran fastStructure from K=1 to K=8 for simple prior.

When using the chooseK.py I get:

Model complexity that maximizes marginal likelihood = 5

Model components used to explain structure in data = 1

I haven't seen cases in the literature where components => "no structure" and complexity => "structure". While I run for logistic prior... any thoughts? Can I conclude the optimal K is between 1 and 5 (not very useful)?