Obtaining harmonized embeddings from bridge integration

[jules-samaran]

Hi,

I am trying to use Bridge integration to integrate rna and proteomic data and additionally to the label predictions I'd be interested in obtaining the integrated dictionary representations in the shared latent space of cells from both modalities.
I guess this is the same question as #8639.
What I'm currently doing after having run PrepareBridgeReference and FindBridgeTransferAnchors is:

latents.ref <- extended_reference[["Bridge"]]@data
latents.query <- Embeddings(obj.query[["ref.Bridge.reduc"]])

However, while these both give me 50-dimensional embeddings, they don't seem well integrated at all (based on a UMAP of the concatenation of both). In #6778, @yuhanH mentioned using the anchors and IntegrateEmbeddings to get the dictionary representations but I'm not sure what arguments I should give to the latter. Should I just use IntegrateEmbeddings, giving as anchors the output of FindBridgeTransferAnchors and as "reductions" the names of the low dimensional reduction (pca in this case) done separately on query and ref (they might be different ones like"lsi" and "pca" when working with atac and rna, would that be a problem?).

Thank you,