Is it possible to have overlaping points in both Jaccard and Bray-Curtis but those points be spaced far apart in Unifrac

General Discussion
1

I had some emperor plots where my points overlap in both Jaccard and Bray-Curtis but are far apart on Weighted and UnWeighted Unifrac.

How is this possible?

If the difference between unifrac and jaccard and bray-curtis is phylogenetic distance. How is it possible to have them be far apart if jaccard and bray-curtis use feature counts. If features are the same, they would classify as the same taxa.

2

Hello Leana,

Welcome to the forums! :qiime2:

Yes, it's possible to get UniFrac distances that are much greater than Jaccard and Bray-Curtis because, out of these three numbers, only UniFrac is using the phylogenetic tree. Just like you said! :palm_tree:

I'm not sure I understand this part. Can you talk me through your thinking?

1 Like
3

I guess I am trying to understand how the jaccard and bray curtis methods calculate. If it is just by matching whether asv are identical wouldn't that indirectly consider phylogeny?

5

Oh, I see what you mean.

Maybe?

The FaithPD alpha diversity metric and UniFrac family of beta diversity metrics are unique because they explicitly reweight their results based on a provided phylogenetic tree.

None of the other metrics use a tree at all.

"matching whether asv are identical" is binary Jaccard, but that does not consider phylogeny directly or indirectly.

I suppose if two samples have identical counts of identical taxa then the Jaccard distance between them would be zero... and their phylogeny would also be identical! :evergreen_tree: == :evergreen_tree:

But phylogeny is not being considered in this calculation. Unless UniFrac or FaithPD is used.

The formulas for Jaccard and Bray-Curtis are listed on this page, if that helps:
https://search.r-project.org/CRAN/refmans/vegan/html/vegdist.html

I feel like I'm missing something. :thinking:

Would you like to share an example to help me understand?

1 Like
6

In addition to excellent explanations from @colinbrislawn, I can only assume that samples that do overlap in Bray-Curtis and Jaccard plots have only a small portion of ASVs that are different between them, with more or less similar abundances but great phylogenetic distances between them since Bray-Curtis and Jaccard doesn't care about how similar or distinct features are; they are either the same sequences or not. So, this small portion of differential ASVs fails to separate samples in non-phylogenetic metrics, but due to the great phylogenetic distances, it causes them to diverge in UniFrac plots.

1 Like