Multimodal data

[grst]

Does sfaira already deal with multimodal data in some way, such as

• RNA-velocity
• VDJ information
• CITE-seq?

[davidsebfischer]

We are preparing for this, in parallel to cellxgene schema 2.0 release. By point:

• CITE-seq will be handled in the same data matrix but the features will be declared separately most likely.
• RNA velocity will be covered via separate layers most likely.
• VDJ will likely be obs columns, so we have to decide which ones we want to streamline. If you have wishes wrt your experience in scirpy, let us know!

Importantly, all of these can already be loaded in the load function, but they will not be considered during streamlining until these features are added. So we add new supported modalities into existing loaders easily.

[grst]

Sounds good, I will definitely check out the cellxgene schema.

w.r.t. the VDJ columns, I would recommend to name the columns as defined by the AIRR rearrangement standard. The columns that are most useful for VDJ analyses and that I import into scirpy be default are

"productive",
"locus",
"v_call",
"d_call",
"j_call",
"c_call",
"junction",
"junction_aa",
"consensus_count",
"duplicate_count",

(see _io.py#L36-L47)