TransferData vs MapQuery vs ProjectUMAP

[royfrancis]

What's the difference between TransferData(), MapQuery() and ProjectUMAP()? They all sounds similar to me. Can someone please explain?

[andrewwbutler]

Hi,

They are all functions used in "reference-based" analyses. Here is how they relate. TransferData is used to transfer categorical or continuous data from a reference to a query dataset - for example to classify query cell types or impute missing features. In theory, you could use the same approach to transfer UMAP coordinates but the UMAP developers have an existing strategy for placing new points into an existing UMAP embedding that is more appropriate (see here). ProjectUMAP is a basically a wrapper to enable that functionality. Often when running these reference-based analyses, you want to do both (e.g. predict cell type labels and generate a unified UMAP plot) and so MapQuery is a higher level wrapper around these both.

[royfrancis]

Thanks for the explanation. A follow-up question. So, as I understand, ProjectUMAP() is purely for UMAP. With TransferData() and MapQuery(), is it possible to use any projection (PCA, ICA etc..)? Or are there any limitations?