Comparison of within-group distances between groups

Hi all!

I want to compare within distances between groups after beta diversity analysis. For example, there are two groups - A and B. I want to show that samples in group A have a more similar microbial composition than that of samples in group B. I could extract distance data of "A and A" and "B and B" from PERMANOVA results. Group B shows higher values than that of group A.
In this case, are there any specific methods available in QIIME2 to compare within-group distances between different groups of samples? Can I just use a general statistical test using these values like the Mann-Whitney test?

Thanks!

Hello Geongoo,

Welcome to the forums! :qiime2:

Which one of these are you asking about?

  • mean(A-to-A distance) vs mean(B-to-B distance)
  • mean(A-to-A distance) vs mean(A-to-B distance) (this one, right?)

A-to-A vs A-to-B is beta-group-significance
A-to-A vs B-to-B is beta-group-significance --p-method permdisp

And what is 'permdisp' :thinking:
https://uw.pressbooks.pub/appliedmultivariatestatistics/chapter/permdisp/

2 Likes

Hi @colinbrislawn

Thanks for your reply.

I want to know the first one. So I want to show group A shares more features between samples than group B. Looks like 'permdisp' is a good option in my case, right?

1 Like

That's correct!

(It's a subtle difference and simple English is ambiguous :upside_down_face: )

At the risk of making this more confusing, here's another way to say it:

The binary Jaccard distance reports the fraction of features that are found in only one sample.

So if all features are shared, then binary Jaccard = 0.0
If 80% of features are shared, then binary Jaccard = 0.2

The goal is to show that:
the mean binary Jaccard distance between samples within group A
is not equal to
the mean binary Jaccard distance between samples within group B

mean(binaryJaccard(A-to-A)) != mean(binaryJaccard(B-to-B))

And the permdisp does exactly this!
(Is this helping at all??)

Thank you so much! It is exactly what I want to do.

1 Like