Longitudinal pairwise-distances on a single group?

I would like to compare microbiome diversity pre- and post- administering antibiotics to a single group of 17 animals.

So far this has worked fine for alpha diversity, using q2-longitudinal pairwise-difference comparisons. However, I've hit a bit of a road block for beta diversity, as q2-longitudinal pairwise-distance comparisons requires a grouping column - which I am unable to provide as my samples are from the same population and treated the same across the board.

I may be missing something conceptually, but is there a way to get around the grouping? To be able to compare if taxa post-antibiotic treatment are significantly different to taxa pre-treatment overall, while keeping the test paired by individual? Or an alternative test available in QIIME2 that allows for the comparison I'm trying to make?

I'm using qiime v.2023.7

Any help is much appreciated,
Best wishes
Dan

Hello!
Having only one group without control group and two time-points (pre and post) is one of the most inconvenient designs for beta diversity (IMHO).

It is pretty easy for alpha diversity, since one may use dependent t-test or dependent Wilcoxon test (depending on data distribution) to test if there is a difference between two time-points. As I understood you already did it.

It is more difficult for beta diversity since beta diversity is expressed in distances, and in your case distances between time points and between groups you want to compare are the same, so you have only one group and nothing to compare with. And performing Permanova test on paired samples will violate assumptions.

There are some workarounds you can try:

  1. Use Mantel test as duscussed here:
    Analysing Paired Data Taken at One Timepoint - #14 by macrobiome

  2. Run Adonis test outside of Qiime2 with SubjectID as strata

  3. Use Adonis in Qiime2 with formula: "SubjectID + TimePoint".

Im not a statistician and not 100% sure about options 2 and 3. Please (all who is reading) correct me if they are not appropriate in that case.

Best,

2 Likes

Hi Timur,

Thanks very much for your response! I agree it's not ideal, but it wasn't possible to collect controls unfortunately.

I'll try using the Mantel test as you suggest.

Best wishes,
Dan