independent samples in alpha and beta diversity

I had an experiment with animals in which 6 groups were fed with addition of different compounds (study groups), and control group that got the same feed with no additives.
Study design: Samples were taken from guts (animals were killed) after 1st, 3rd and 5th week of experiment.
So I have 168 samples in following 21 groups with 8 samples in each:
control group after 1st week n=8, control group after 3rd week n=8, control group after 5t n=8,
group fed with addition of compound A after 1st week n=8, after 3rd week n=8, after 5th week n=8 , group fed with addition of compound B after 1st week n=8, 3rd week n=8 and so on...
And I would like to emphasize that animal in first week of experiment is not the same animal as in 3rd and 5th week in one group (animal were killed before each sampling). So I have independent samples because they were derived from different animals.

I want to compare samples from different time points (1st 3rd 5th week) within one group, and according to the each week in cross-sectional way. I mean compare eg. each study group in 1st week to control group in 1st week and so on.

My question is: can I perform alpha and beta diversity like it was showed in Moving picture tutorial: an analysis of human microbiome samples from two individuals at four body sites at five timepoints (“Moving Pictures” tutorial — QIIME 2 2022.2.0 documentation) ? Are these samples dependent?
While Kruskal Wallis test used in alpha diversity takes only independent data. So in moving picture tutorial these samples were treated as independent?

Sorry for long explanation but I wanted to be understood in a proper way:)

Hello Ewelina,

Thank you for the detailed description of your study. This sounds like an interesting cohort and you have a lot of groups to compare!

You have identified the core issue; due to destructive sampling, the change of timepoint is confounded by the change of animal.

We just discussed what to do with destructive sampling in this thread. You may find it useful:

I again recommend qiime longitudinal first-distances with --p-baseline which is fine for destructive sampling.

Some methods work best with fully independent samples (like you have!) and some work better with matched samples. :person_shrugging:

If you have not yet asked a statistician in your department, now is a great time to do so.


Hi @colinbrislawn !

Thanks for your answer and useful link, I will do that.
I haven't ask any statistician yet, and yes I should.


1 Like