Use spatial autocorrelation for host phylogeny?

[maureenbaars]

Hi,

I’m analysing microbiome data from different host species with gllvm models. I now model with host species as a random effect, however I’d like to incorporate host species phylogeny in the model instead. Would it make sense to incorporate host phylogeny in the same way as spatial autocorrelation, for example by using a covariance matrix of species instead of a distance matrix with coordinates?

The model I’m currently using:
model_sp <- gllvm(asv_table, studyDesign = sDesign, family = "negative.binomial", row.eff=~(1|host_species), num.lv = 2, sd.errors = TRUE)

Many thanks in advance!
Maureen

[BertvanderVeen]

Thanks Maureen. Have you had a look at our vignette on phylogenetic ransom effects? See https://cran.r-project.org/web/packages/gllvm/vignettes/vignette7.html and also https://arxiv.org/abs/2408.05333.

[maureenbaars]

Hi Bert,

Thanks for your response. I have looked at this vignette, but from what I understand, this is about adding the phylogeny of the input species (but please correct me if I’m wrong). I’m analysing the composition of bacterial taxa (the asv_table in my model) sampled from different hosts from various species. As I expect the bacterial taxa composition to be more similar in more closely related host species, I’d like to incorporate host species phylogeny into the model. How could I do that?

Thanks,
Maureen

[BertvanderVeen]

Ah, I see! There is currently only one way to incorporate phylogeny in gllvm; that is for the column identities in the response variable. If you want to do something else, you could have a look at the glmmTMB package which is more limited in that it can only do phylogenetic random intercepts, but that corresponds with what you want here!