How to measure the strength of a correlation between two sample sites

[mweinroth (Maggie Weinroth)]

I am looking for the best tool to provide a paired comparison of beta diversity between one sample site of interest and 9 'preditors'.

My experimental setup is as follows: 10 animals were each sampled in ten sites in the GI. I would like to know which site is most similar to the last site (swab). Ideally, I would like not only a p-value type statistic but also something like an r-squared value so I could say something like "site A can explain 37% of the variation in the swab while site B can explain 50%".

I have had a few ideas but am not sure which direction to go:

• Use qiime diversity adonis as suggested here however this will not be quite right because if I leave one sampling type and the swab it will just tell me the variation explained the site not the similarity between the two.

• Use q2-longitudinal but from the tutorial it seems like I only get a p-value with no R2.

• Treat my swab like a mock community and compare my other sample types individually to it like here but this seems not really used and outdated

• Use Procrustes and the m2 statistic as an R2

Are any of these more or less desirable than the other or am I missing a better option?

Thanks!

[Nicholas_Bokulich]

Hi @mweinroth ,

It sounds like what you want is to inspect the distributions of beta diversity distances/dissimilarities between each group. What you want is not an R2 value, but the distances themselves (e.g., report median distance).

Pick your distance metric(s) of interest and use qiime diversity beta-group-significance to generate a boxplot showing the distribution of distances within and between each group.

That action can also report p-values for pairwise comparisons of each group.

@mweinroth

Treat my swab like a mock community and compare my other sample types individually to it like here but this seems not really used and outdated

interesting idea! But the information retrieved is rather specific (which is probably why it is used less often).

@mweinroth

Use q2-longitudinal but from the tutorial it seems like I only get a p-value with no R2.

that would work! the R2 value would instead be inspecting the mean or median distance values themselves.