I would like to perform pairwise comparison on single pair of samples. This is, if I have 7 samples from S0 to S6, compare them as indicated below to see how much the microbial communities differ between them:
S0-S1
S0-S2
…
S0-S6
The samples are from the same ‘environment’ but they have been collected at different time points.
I am unsure which is the best test for doing this. Maybe pairwise difference comparisons using unweighted unifrac pcoa results (however I would just have one single sample in each group)? Or what would be the most appropriate option?
My recommendation would be to calculate beta diversity! (This is actually the defination of beta diversity). You can either do this with the core-metrics pipeline, or directly work with a rarified table.
As far as statistics go, there’s not a lot you can do here. You can look at the spread of the sames, and try to determine whether some are similar but without additional metadata, its kind of hard to classify them.
Yes, I have run qiime longitudinal first-distances (both with and without baseline) to asses how the rate of change (of the beta diversity distances) differs over time in my sample set.
However, I am more interested in particular features that change over time. For that, I have run qiime longitudinal feature-volatility. The problem is that given my low number of samples and the individual-specific differences I have between my two subjects in the analysis, the model is inaccurate, then I guess that the features are meaningless. My idea is to just run it with a single subject, but I am unsure if that is appropriate or not…
Based on what I’m hearing, you have an answer: you’re probably underpowered to do this analysis. I can’t give you a better power calculation, but at 2 subjects, finding features seems… optimistic all things considered. So, my suggestion would be say that you haev a difference in beta diversity but dont have the power to detect single features over time.