Absolute counts or relative abundance for BC distance matrix

[MonicaSteffi]

Dear everyone,

I have a dataset of absolute counts and relative abundance obtained from qPCR experiments. I'm interested in calculating the Bray-Curtis dissimilarity to analyze the community structure across different samples.

I plan to use the vegdist function from the vegan package in R.

My main question is:

Can I use relative abundance data directly with vegdist, or do I take the absolute counts abundances to calculate Bray-Curtis dissimilarity?

I'm particularly interested in understanding if there are any pitfalls or considerations I should be aware of when using relative abundance data for this type of analysis. Any insights or recommendations would be greatly appreciated!

Thanks in advance!

[gavinsimpson]

What are you interested in modelling? Abundance or compositional difference among levels of the treatment?

Bray Curtis is not a good choice for either of these; the resulting dissimilarity includes a component due to abundance (size) differences between the pair of samples, and a compositional (shape) component. This is problematic if you want to estimate abundance difference or compositional differences as the dissimilarity contains both.

You can use either - it's just a mathematical operation on the data - but you should use the counts (possibly transformed) if you really must use it, as this will be as close to the what it was intended for. However, it's going to give you a test that conflates abundance (size) and composition (shape) differences.

Why do you even need to convert your data to dissimilarities?

[MonicaSteffi]

Why do you even need to convert your data to dissimilarities?

I have bacterial counts at different time points for particular experiments. I just want to PCOA plot at the end to know whether microbial communities are different with different time points.

[gavinsimpson]

PCoA won't tell you that - you'd need dbrda() a Distance-based RDA for that, and you wanting to do a PCoA isn't an answer to why you want dissimilarities. You are throwing away information by converting to dissimilarities and you could avoid this through PCA/CA or RDA/CCA (depending on what you want to look at) just fine. I know that dissimilarities is how everyone seems to want to analyse these molecular data but there is little good reason in general to do so, other than it's what everyone does (which isn't a good justification).

[jarioksa]

This is not a vegan issue: you should consult your scientific community. In general, relative counts are OK and Bray-Curtis may not be optimal if you have very varying totals. (Neither is RDA/PCA in that case.)