If I do PCoA on a dataset, can I use these scores in subsequent analyses? My understanding is that Using PCA scores in subsequent regression is valid. However, it seems like this doesn't hold for NMDS scores. However, it seems also seems like PCA is not appropriate for my data sets which are zero-inflated.
To give an overview of the problem I'm working on, I have two large matrices of species composition - one that contains plant community composition (~150 variables), and one that contains soil bacteria (~11,000 variables), each for 28 data points. I would like examine how well we can predict a univariate response (soil carbon storage) from these community matrices. So broadly, I'd like to simplify each of these community matrices and use the resulting axes as predictors either in some kind of linear model or variance partitioning analysis so that I can compare their usefulness as predictors. Can I do this with PCoA scores, or should I just use PCA scores? What would be the benefit or drawback of either approach?
