Other visualisation methods

[sa-lee]

Since structure model is the same thing as LDA (with 'documents' being samples, 'topics' as latent populations, and 'words' as genetic markers) we could use some of the ideas in
http://nlp.stanford.edu/events/illvi2014/papers/sievert-illvi2014.pdf
https://github.com/cpsievert/LDAvis
to create some more informative/compelling plots

See also https://en.wikipedia.org/wiki/Latent_Dirichlet_allocation

[sa-lee]

• plotMDS - mds plot using Jensen-Shannon metric or Structure output
• plotRanks - bar plots ranking SNPs/microsatellites by cluster (need some kind of ranking meetric here)

Then our fully interactive plot that is like LDAvis

[sa-lee]

Other ideas:
clustering tree ala http://lazappi.id.au/building-a-clustering-tree/
including uncertainty estimates in Q matrix from admixture