Rotating the view and making an animated .gif or similar seems to work quite well in many instances.

One way to interpret the plot might be that PCA is representing everyone as admixed between three idealized populations

this is not really right, but goes in the right direction. don’t confuse this for mode-based clustering, which does assume ‘pure’ populations in many cases. the other dimensions are in the results.

Thoughts? Investigations?

What I want to do here is to visualize the scatter but allowing to shift positions to geographic coordinates (by interpolating). I will do this using an interactive scatter that allows user to focus on regions, similar to what I’ve done here: http://moebio.com/research/wikipediagender yet allowing zoom in any chosen coordinate

Other interesting approach is to draw inverse geographic distances with lines (geo closeness network)… and the opposite, draw the genetic closeness network when point are placed on geographic positions.

Finally it would be fantastic to use PC1/PC2 coordinates to place populations and then perform a continual deformation of the geographic coordinates space in order to make the regions cover their population spots. That would be a new genetics-driven geographical projection. I love space deformations a la D’Arcy Thompson or a la Einstein/Poincaré and I’ve already worked on some deformed maps, look: http://moebio.com

What do you think of these ideas?

https://www.dropbox.com/s/ar7d4u6izbexq3v/plinkmds.csv

https://www.dropbox.com/s/eqvuy11h2kgy9bm/hgdphapmapbehar.csv

i assume most of you know R, but useful:

attr(kVert,”names”)

inspect them and remove populations you don’t want to plot (e.g., kVert[22]=NULL). “friends” you should remove, they’re a bunch of random people (my readers & friends) who create a very large circle because they’re not a real population.