Converting `Seurat` fails when graphs don't have a `DefaultAssay`

[rcannood]

I encountered a Seurat object with the following content:

An object of class Seurat 
73202 features across 342178 samples within 2 assays 
Active assay: RNA (36601 features, 2000 variable features)
 4 layers present: counts.10x_3p_v3.1, counts.10x_5p_v2, data.10x_3p_v3.1, data.10x_5p_v2
 1 other assay present: sketch
 5 dimensional reductions calculated: pca, harmony, umap, harmony.full, umap.full

---

Calling as_AnnData() fails:

> adata <- as_AnnData(data, assay_name = "RNA")
Error in if (SeuratObject::DefaultAssay(graph) != assay_name) { : 
  argument is of length zero

On this line:

anndataR/R/from_Seurat.R

Line 464 in 1d9145e

if (SeuratObject::DefaultAssay(graph) != assay_name) {

I can't share the data file, but I notice it has two graphs. I can read the graph just fine, but the assay.used seems to be empty for some reason, thus causing the exception above.

> SeuratObject::Graphs(data)
[1] "sketch_nn"  "sketch_snn"
> gr <- data[["sketch_nn"]]
> gr
A Graph object containing 55331 cells
> gr@assay.used
character(0)
> SeuratObject::DefaultAssay(gr)
NULL

I suppose that in this case the graph could be skipped, but it would be nice if as_AnnData didn't throw an error but a warning instead

---

In this case, I'm able to circumvent the issue by turning off the obsp_mapping for now until this issue is fixed.

adata <- as_AnnData(data, assay_name = "RNA", obsp_mapping = FALSE)

[lazappi]

Hmmm... I wonder how that happened? Maybe someone created a graph manually or it comes from an old version of {Seurat}...

Either way, this should be handled with a warning. I'm not sure if it's better to skip it or convert it anyway though?