Bray curtis PCoA vs UniFrac interpretation

Hi everyone,

I need some help to interpret the beta diversity results better. I am especially confused by Bray-Curtis results when compared with weighted, unweighted and taxonomic analysis.

In Bray-Curtis, I observe a clear clustering pattern by library preparation plate: samples from plate 1 cluster together in cluster 1, whereas those from plate 2 and 3 cluster together in another cluster. Once I look at the taxonomy, I am not detecting any specific taxonomic profile driving this divergence. Moreover, I am not able to see the pattern in beta diversity UniFrac metrics (weighted or unweighted).

Bray-Curtis PCoA colored by library plate:
image

Weighted UniFrac PCoA colored by library plate:
image

I feel that same bacterial species are classified to different ASVs, maybe driving this divergence in Bray-Curtis, which is lost when using phylogeny-aware metrics.

How should I interpret these results? Do we have a technical bias in this dataset? Why is this pattern not observed in UniFrac metrics? How can we identify the specificities of this issue?

Thanks a lot for your time.

Best,

Anna

1 Like

Hi @AnnaC,

I think your interpretation is on the right track:

With Bray-Curtis, you weight the difference between two ASVs that differ by a single nucleotide with the same weight that you’d place on two ASVs that differed by 100 nucleotides, where as the UniFrac distances are able to account for the fact that the two ASVs that differ by one nucleotide are much closer. So, my interpreation would be that you have a difference in closely related ASVs between the orange and blue/red runs.

One question is whether your bioinformatic parameters changed. I would double check how you built your tables before they went into beta diversity. Make sure your parameters are the same because that might explain why you’re getting similar but slightly different ASVs between the two runs.

Best,
Justine

2 Likes

Thanks, @jwdebelius!!!

I think it might be this. The ASVs, which are giving the difference among clusters, are identical in the exception of a single base in the beginning and/or in the end. We are trimming the primers by length since we do not have the sequence (commercial primers), so I think the issue might be there.

1 Like

2 posts were split to a new topic: Trimming ASVs in table without reprocessing