PDA

View Full Version : Plotting of PCAs



mephisto
02-14-2018, 12:01 PM
Depending on what PCs (Principal Components) I assign for the X- and Y-axis the positions of the various samples change. How can you make a plot where all 20 Eigenvectors of a person are considered? Scatterplots where only 2 PCs are considered are only a partial truth since 18 other eigenvectors are completely unused. Is this possible?

Tolan
02-14-2018, 04:54 PM
I think, the solution is to assign each component a latitude and a longitude value.
And then multiply the percentage of the calculator of the sample with this value, and then do the average.
That's what I did for Eurogenes K15. http://gen3553.pagesperso-orange.fr/ADN/K15.htm
I calculated the latitude and longitude for each component, averaging from the calculator results, from a few well-distributed countries on all continents.

I reported some components in this map (the black squares)

Sangarius
02-14-2018, 05:03 PM
Depending on what PCs (Principal Components) I assign for the X- and Y-axis the positions of the various samples change. How can you make a plot where all 20 Eigenvectors of a person are considered? Scatterplots where only 2 PCs are considered are only a partial truth since 18 other eigenvectors are completely unused. Is this possible?

Well, you obviously can't represent 20 dimensions in a two-dimensional scatterplot. Try clustering methods if you want a visualization that takes all dimensions into account.

Volat
02-15-2018, 05:55 AM
Unless you have large dataset usually first 2 principal components will account for most variation in data. Above 80% or even 90% . If that the case, then you don't need to bother. Otherwise cluster analysis and dendrogram as recommended above.

lukaszM
02-15-2018, 08:11 AM
Unless you have large dataset usually first 2 principal components will account for most variation in data. Above 80% or even 90% . If that the case, then you don't need to bother. Otherwise cluster analysis and dendrogram as recommended above.

Or make 3D PCA plot which use three components. Even fourth component could be visualized on 3D plot, as size of bubbles ,and fifth as color of them.

Volat
02-15-2018, 09:33 AM
Or make 3D PCA plot which use three components. Even fourth component could be visualized on 3D plot, as size of bubbles ,and fifth as color of them.

Sixth component as intensity of the colour? But will you be able to see underlying structures in populations using such constructs?

lukaszM
02-15-2018, 09:35 AM
Sixth component as intensity of the colour? But will you be able to see any patterns in such constructs?

I tried some time ago with 5 components. Very hard to deal with all of them at once. Our brain isn't accustomed to visualize in more than three dimensions...