How to use pcoa-biplot


I am really interested in qiime diversity pcoa-biplot.
In my understanding, this plugin can show the features (with taxonomic infromation?) with arrow on PCoA space and help us to understand the relationship between features and variance of plots. Am I right?

But I could not understand the usage of this plugin well.
This plugin requires FeatureTable[RelativeFrequency].
Does anyone know how to prepare the FeatureTable[RelativeFrequency]?
And if it is possible, I would like to visualize taxonomic information upon PCoA emperor plot.

Thanks in advance.


Hi @suzukik!

Great question, there’s a few steps to this:

  1. Use qiime feature-table relative-frequency to generate the table for pcoa-biplot. You need this to be based on the rarefied table you are using for your beta and pcoa steps. You can use provenance to prove that you’re using the same rarefied table in both relative-frequency and pcoa :slight_smile:

  2. Run pcoa-biplot with that table and an existing PCoAResults (generated by your previous pcoa).

  3. Use qiime emperor biplot with --m-feature-metadata-file set to your FeatureData[Taxonomy] artifact. This will let you color the biplot vectors by taxonomy (it’s at the bottom of the normal color dropdown in the Emperor UI).

Abstractly the pipeline looks like this:

`rarefy` ---> `beta` -> `pcoa` -> `pcoa-biplot` -> `emperor biplot`
       \-> `relative-frequency`--/                /
                                   taxonomy.qza -/

Hi @ebolyen,

Thank you very much for your great support!
I got the picture as bellow.

I would like show the taxonomic name on the arrow instead of featureID. If it is possible, I'd like to show more simple name like g_XXXX;s_XXX.
And can I export pcoa scores (contributions) of feature IDs (with their taxonomy, if possible)?

Thank you for your cooperation.

Hi @suzukik,

Glad it worked!

You will probably want to use qiime taxa collapse (unfortunately that has to be done of the frequency table). To my knowledge emperor does not let us rename the vectors you see there, so we have to provide a different feature set (via taxa collapse, or feature-table group).

It sounds like you are interested in something like the loadings/coefficients of each feature for a sample? Unfortunately that is only possible with Euclidean distance (PCA), with general metric spaces we don’t have that option.

However if you are interested in the specifics of the feature vectors in terms of the principle coordinates, then you can export your PCoAResults % Properties("biplot") artifact to see that (it’s a TSV with a few sections). But my guess is that isn’t probably what you are looking for.

cc @yoshiki

1 Like

Dear ebolyen,

Thank you very much for your support. And I’m really sorry for my late replying.

OK, I understand that it is difficult to rename the vector in emperor plot.
Is there any option to hide the vector name (show just arrows)?

About 2nd question, I successfully extracted the results of PCoA with scores of features.
Thank you very much.

Best regards

Hi @suzukik,

I am similarly sorry for my late response,

I do not believe so. Although if you were to export the plot as svg (via a right-click), you could edit out the labels manually. Obviously it won’t be interactive anymore, so that may not be what you need.

Hi @ebolyen,

Thank you for the replying.
Yes, you are right. I will edit them manually.
Thank you for your kind support.


This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.