Features in Principal Coordinate Analysis Biplot

timanix · July 2, 2019, 10:52am

Dear all!
I got nice biplot visualizations for my data. Could you explain me by which criteria features, represented in the biplot were selected?
When I extracted data for arrow plotting from biplot.qzv files and took first features sorted by the biggest absolute values I received in some cases the same features, but in other different.

thermokarst · July 2, 2019, 9:36pm

Hey there @timanix!

I had to dig a little bit here (which means we could probably do a better job documenting this):

github.com

qiime2/q2-emperor/blob/0050dd5f882909bdd7bf2d58dbf8a8ed065de5c2/q2_emperor/_plot.py#L84-L90


      
          # select the top N most important features based on the vector's magnitude
          feats = biplot.features.copy()
          origin = np.zeros_like(feats.columns)
          feats['importance'] = feats.apply(euclidean, axis=1, args=(origin,))
          feats.sort_values('importance', inplace=True, ascending=False)
          feats.drop(['importance'], inplace=True, axis=1)
          biplot.features = feats[:number_of_features].copy()

So, it looks to me like the "importance" is calculated for each feature based on the vector's magnitude (euclidean distance from origin). The n largest "importances" are then plotted in the emperor viz (5 largest, by default).

Hope that helps!

timanix · July 3, 2019, 6:46am

Thank you a lot for your answer.
This code is exactly what I needed!!!
Now in the scatterplot I can plot the same features as in biplot.qzv! Perfect!