beta diversity - pairs of samples

Hey,
How do I interpret beta diversity between individual samples? (all tutorials are between group of samples)
I’ve used three metrics correlation, jaccard, braycurtis
ie:

qiime diversity beta \
 --i-table table-rarefied.qza \
 --p-metric jaccard \
 --o-distance-matrix beta.jaccard.qza

the example results:
jaccard:
image

bray-curtis:
image

correlation:
image

what is the meaning of those numbers? Why Correlation is 0 in the same samples?

For example s10SWB, s10SW, s11SW cluster together in PCoA, as indeed those are the samples from the same guy. In Jaccard they have values ranging from 0.67 to 0.76 - so should I interpret it that “similar numbers” have similar diversity? Similar low? Similar high?

Maybe there is some webpage saying Jaccard:
0.0 - 0.2 - not similar diversity
0.2 - 0.4 - a bit similar diversity
0.4 - 0.6 - a bit more similar diversity
etc…

Thanks for any help/hints

These numbers are distances.

small distance == close == similar
larger distance == far away == different

This is why samples have 0 distance from themselves; they are identical == zero distance from themselves

Colin
P.S. Technically bray-curtis is a dissimilarity. :man_shrugging:
P.P.S. “all tutorials are between group of samples” You have got to have reps for stat tests!

Normally correlation means: 0 - no relationship, 1 - full relationship (regards Pearson/Spearman). So at least the “correlation” beta metrics is not clear to me.
Anyway, if 0 is “no distance” than it’s strange for me that repetitions from the same patient are such distant (but in PCoA they are close)

Let’s see if we can find the official docs about how these are calculated to double check!

Any tips @thermokarst !

1 Like

Hi @thedam,
This page explains all of the b-diversity metrics available.

Correlation distance is 1 - correlation coefficient

2 Likes