3D vs 2D emperor?


I wonder why the emperor figure is 3D? Can't I control that to be 2D and still run the statistical test in qiime2?
Here my emperor if you would like to have a look:
bray_curtis_emperor.qzv (861.7 KB)

Many thanks

Hi @MarwaTawfik

If you will click "Axes" on the right part of the page, you can select only 2 axes to show (just hide a third one), so it will be 2D. But 2D or 3D is not important for stat analysis, since it is only related to the visualization.


Thanks very much Timur @timanix
How do to decide on which one to use (2D or 3D)?
I think it is difficult to visualise 3D in papers
If the differences are not clear in the 2D, it is better to use the 3D option?

One last query if you don't mind, could you please advice if there is an option to circle each group/subgroup on the figure itself?

Just choose axes that are best explaining your data.
3D can help with finding best axes, but usually first 2 are the best.
If separation is not clear, stat. test as PERMANOVA or others will tell you if there are differences in the distances


Unfortunately, this functionality is not implemented, so to do it like this you may need to use some R or Python libraries. You can export Qiime2 data and perform visualisation with other tools/libraries.


Hello again, Timur @timanix
Many thanks again.

Any way to only choose certain groups out of these for visualising
I tried different tabs in view.qiime2 but can't find away.
Do I need to rerun qiime2 on certain subgroups for this purpose?


Hi @MarwaTawfik!
Actually, you have several options:

  1. The simplest one (not recommended!). There should be a tab called something like "visibility" (do not remember exactly, answering from the smartphone), were you can set visibility of certain groups to 0.
  2. If you need not only to hide groups from the visualization, but also to remove then from the distance matrix, you can filter them by metadata.
  3. You also can filter your feature table and recalculate diversity metrics.

Hi @timanix
Thanks very much
So for filtering (either metadata or feature table based_ I would go for

But do you think I'd need that? does removing samples/certain groups have any bad effect on visualisation? not sure if that question need to be directed to the developers themselves

Personally I would filter already calculated distance matrices and recreate PCoA plots, since all groups you will hide already influenced ordination of the dots on the plot. This way you will get the best clusters.
If you will filter feature table and recreate diversity metrics, it would be slightly different since samples in your filtered rarefied table, rarefied to the same depth, will still be different from the same samples in previous table (due to the randomness of rarefaction process).