FindTransferAnchors => allow mapping of features between ref and query

[nick-youngblut]

All of the features in my query are fully uppercase, while my reference features are capitalized. While mapping between the 2 character vectors is easy: rownames(query_obj) == toupper(rownames(ref_obj)), renaming features in a seurat object is not trivial, so it is challenging to get past the No features to use in finding transfer anchors generated by FindTransferAnchors.

It would be helpful to FindTransferAnchors included a parameter that allowed users to map feature names between the query and reference (similar to the dplyr::join_* functions). Currently in Seurat v5, there is the features parameter in FindTransferAnchors, but this just subsets from the total possible features, which still must match between the query and reference.

[kmshort]

Just a guess, but are you mapping human to mouse of vice versa? If so, I am doing that now. I'm doing the reverse, human to mouse, but you might need to consider checking for orthologous genes because not all these species' genes are the same nomenclature. I've grabbed all mouse ortholog names (to the human genes) for integration and transfer, and left the remaining human genes that have no representation in mouse alone. The human-specific genes will be no good for anything like DGE comparisons, but their presence still helps with PCA, scaling and dimension reduction within its own dataset/layer. Anyway, using "toupper" is not the appropriate way to go IMO. While many gene names will simply be changes in capitalisation, some gene names will not. I've used biomaRt to do it, but it's a pain and ensembl seems to be timing out for me at the moment, so I downloaded a human<>mouse ortholog gene list directly from biomart via the web as a csv and reference it directly.

[nick-youngblut]

@kmshort it's mouse to mouse, but the captilization varies.

using "toupper" is not the appropriate way to go IMO

I agree. It would be very helpful if Seurat allowed for flexiblity with feature naming and metadata, given that Ensembl accessions (or similar) should really be used for feature mapping in order to provide a consistent mapping across datasets, instead of gene names, which can vary; however, researchers ultimately want to use gene names and not accessions. So, it would be helpful if one could seemlessly transition back and forth between different feature naming (e.g., the actual feature names versus columns in the feature metadata).