modeling bacterial features as indepedent variables

[jwdebelius (Justine Debelius)]

Hi @kam,

@kam

. How would you approach this if, for example, each participant has two samples (time 1 and time 2) and you would like to model the change for each feature as the independent variable in the regression?

This is more of a question about what kind of model do you use for repeated measurements? I tend to use a linear mixed effects regression for this. You could theoretically also use a GEE; you should consult your local statistician for the proper way to do this modeling. Personally, it took me a little bit to work through the whole repeated measures thing (there was a like two hour phone call with a friend who's a statistician where he was extremely patient with me). I'm like 85% sure that ANCOM-BC either has or will soon have an LME component, so I'd probably recommend you just go for that.

I'm going to link you to another post that come up in the past day or two on longitudinal analysis. Full disclosure, it has some paired sample analysis I did for the project which includes a polymicrobial microbiome measurement I used as a a predictor in an odds ratio model. (My epidemologist was so happy with me!)

pairwise differences on Axis 3

Hi @nathaniel_hubert, I think rather than the biplot and correlation you might want to look into either complex tensor factorization in Gemelli or rPCA in DEICODE. The advantage of these over traditional metrics is that the features are embded in the ordination, so you can use them to figure out what side of the ordination features are associated with... or to even build ALRs if you're into amalgamated microbial statistics. I happen to really like amalgamated values If it's helpful or interest…

Best,
Justine