Entering edit mode
3.7 years ago
Aspire
▴
360
I think that the axis of a PCA plot represents the distances between samples - the larger the difference, the larger the scale will be.
Does the location of the samples relative to zero matter?Is there a larger difference between two samples, located at 5 , -5 on one of the axis, than between another two samples located at 5, 15?
For example, can one say that samples 1 and 3 are more dissimilar than 2 and 4 in the plot below?
Don't think the value of PC can be directly translated into distance. What you want is the IBD / IBS calculation to assess the distance between samples.
On the other hand, if you are interested in population distance, then you can do something like this
In the direction of PC2, 1&3 have a distance of 100 units, as do2&4. Yes it's linear. However, in the direction of PC1 we cannot say from this picture, and PC1 is a larger share of the raw data than PC2. Dont do distance in units you don't understand, check distance in the original dataset.