using phyloseq distance with normalisations from DESeq or edgeR

[FabianRoger]

Hej,

sorry for this naive question, but after reading your convincing article ("waste not want not") I would like to use the proposed techniques on my data. I especially want to calculate distance matrices with the appropriate alternatives to rarefaction but I am not sure how and if this has been implemented in the distance() function from the phyloseq package. It doesn't seem to have an option to choose the normalisation method (e.g. rarefaction, proportions, edgeR or DESeq). I found code for these normalisations in the supplementary but the resulting objects are in the format "DGEList" which is not supported by distance().

Before I spend more time trying to stick together the necessary lines of code, I just want to make sure that I didn't miss the obvious and that it hasn't been implemented in some function that I have missed or in some extension?

Any help would be appreciated!

Thanks for your great contributions,

Fabian

[joey711]

Transformations of the count values will be upstream of the distance() function call, except in cases where the distance method that you select already has a normalization implied/required in the method.

So for your case you would perform the variance stabilization first, and then replace the otu_table component with the variance stabilized version.

dds = phyloseq_to_deseq2(physeq, ~Treatment)
# Make a copy of your phyloseq object, which you will then modify with VST values
physeqvsd = physeq
vsd = getVarianceStabilizedData(dds)
otu_table(physeqvsd) <- otu_table(vsd, taxa_are_rows = TRUE)
distance(physeqvsd, ...)

This should be enough to get you started. Variance stabilization often produces negative values for very rare/low counts. If that is a problem for the distance method you plan to use, you can set these to zero, add pseudocounts before VST, or some other procedure.

If getVarianceStabilizedData is taking too long, you might try rlog, also in DESeq2.

See the following closed issue for more details...

#445