I have a multidimensional feature that represents a 3D vector space obtained from high-dimensional morphological data and reduced with an algorithm like UMAP or tSNE that reduces the dimension non-linearly. I'm thinking of reconstructing the coordinates of this 3D space into a phylogeny. Could you reconstruct coordinates at each internal node of a phylogeny, from the known coordinates at the tips of the tree? In reading this paper and the associated book, I have only seen the multivariate trait method but there are no examples of mapping in the phylogeny. I have the doubt if what I imagine is possible with some method that integrates the coordinates and not only uses a single coordinate like this figure that maps only one dimension of the PCA.