Computing beta-diversity on clr-transformed counts

1

Hi,

I followed the steps for pre-processing of 16S by CLR-transformation of TSS abundance.

If I wanted to compare the PCA with the results for beta-diversity, does it make sense to compute for the beta-diversity (eg, Bray-Curtis) on the clr-transformed abundance? Or should I compute it on the untransformed relative frequencies?

Thank you.

2

hi @ange
First we have updated our guidelines here: http://mixomics.org/mixmc/mixmc-pre-processing/
There is not need to fo a TSS transformation before a CLR transformation because CLR is invariant.

Regarding the Bray Curtis distance, it does not make sense for compositional data. You can refer to this article and in particular Figure 2 that recommends using a CLR transformation with PCA (Aitchison distance).

Hope that helps,
Kim-Anh