Distortion functions on the embedded positions?

[slinnarsson]

Hi - thanks for developing pyMDE!

I am trying to create an embedding like tSNE or UMAP, but with additional bias on the embedded space. For example, I have single-cell RNA-seq data and would like to embed according to cell-cell similarity as usual. In addition, each cell (sample) has an attribute, for example "age", and I would like to bias the embedding to place larger values of the attribute further from the center.

Is there a mechanism to create a distortion function that operates on the absolute positions in the target space? I understand that constraints allow you to pin datapoints to specific absolute positions, but I want a more flexible bias rather than pinning points precisely. Any help appreciated!