Merge pull request #56 from TileDB-Inc/aaronwolen/sc-17824/support-st…

…orage-of-seurat-cell-identities

Support storage of Seurat cell identities

DESCRIPTION

@@ -5,7 +5,7 @@ Description: Store and retrieve single cell data using TileDB and the on-disk
   format proposed in the Unified Single Cell Data Model and API. Users can
   import from and export to in-memory formats used by popular toolchains like
   Seurat and Bioconductor SingleCellExperiment.
-Version: 0.1.0.9009
+Version: 0.1.0.9010
 Authors@R: c(
     person(given = "Aaron",
            family = "Wolen",

NAMESPACE

@@ -23,6 +23,7 @@ importFrom(SeuratObject,CreateSeuratObject)
 importFrom(SeuratObject,DefaultAssay)
 importFrom(SeuratObject,Embeddings)
 importFrom(SeuratObject,GetAssayData)
+importFrom(SeuratObject,Idents)
 importFrom(SeuratObject,Loadings)
 importFrom(SeuratObject,Reductions)
 importFrom(SeuratObject,SetAssayData)

NEWS.md

@@ -34,3 +34,4 @@ See [TileDB 2.8 release notes](https://github.com/TileDB-Inc/TileDB/releases/tag
 * New internal `TileDBURI` class for handling  various URI formats
 * The `uri` field for all TileDB(Array|Group)-based classes is now an active binding that retrieves the URI from the private `tiledb_uri` field
 * Several default parameters have been changed to store the the `X`, `obs`, and `var` arrays more efficiently on disk (#50)
+* Seurat cell identities are now stored in the `active_ident` attribute of the `obs` array (#56)

R/SCDataset.R

@@ -3,7 +3,7 @@
 #' @description
 #' Class for representing a sc_dataset, which may contain of one or more
 #' [`SCGroup`]s.
-#' @importFrom SeuratObject CreateSeuratObject Reductions
+#' @importFrom SeuratObject CreateSeuratObject Reductions Idents
 #' @export
 SCDataset <- R6::R6Class(
   classname = "SCDataset",
@@ -56,6 +56,12 @@ SCDataset <- R6::R6Class(
     #' a nested TileDB group with a URI of `./scgroup_<assay>` where `<assay>`
     #' is the name of the Seurat assay.
     #'
+    #' ## Identities
+    #'
+    #' Cell identities in the [`SeuratObject::Seurat`] are maintained by
+    #' creating an `active_ident` attribute in `obs` that stores the factor
+    #' levels as a character vector.
+    #
     #' ## Dimensionality Reductions
     #'
     #' Dimensionality reduction results are stored as `obsm` and `varm` arrays
@@ -68,6 +74,15 @@ SCDataset <- R6::R6Class(
     from_seurat = function(object) {
       stopifnot(inherits(object, "Seurat"))
 
+      idents <- SeuratObject::Idents(object)
+      if (nlevels(idents) > 1L) {
+        object <- SeuratObject::AddMetaData(
+          object = object,
+          metadata = as.character(idents),
+          col.name = "active_ident"
+        )
+      }
+
       assays <- SeuratObject::Assays(object)
       for (assay in assays) {
         if (is.null(self$members[[assay]])) {
@@ -139,12 +154,20 @@ SCDataset <- R6::R6Class(
       # just take the first scgroup's obs metadata
       obs_df <- self$scgroups[[1]]$obs$to_dataframe()
 
+      # retain cell identities before restoring cell-level metadata
+      idents <- setNames(obs_df$active_ident, rownames(obs_df))
+      obs_df$active_ident <- NULL
+
       object <- SeuratObject::CreateSeuratObject(
         counts = assays[[1]],
         project = project,
         meta.data = obs_df
       )
 
+      if (!is.null(idents)) {
+        SeuratObject::Idents(object) <- idents[SeuratObject::Cells(object)]
+      }
+
       if (nassays > 1) {
         for (i in seq(2, nassays)) {
           assay <- names(assays)[i]

tests/testthat/test_SCDataset_Seurat.R

@@ -1,13 +1,5 @@
-setup({
-  tdb_uri <<- file.path(tempdir(), "test-scdata")
-})
-
-teardown({
-  tiledb::tiledb_vfs_remove_dir(tdb_uri)
-})
-
-
 test_that("SCDataset can be created from a Seurat object", {
+  tdb_uri <- withr::local_tempdir("test-scdataset")
   scdataset <- SCDataset$new(uri = tdb_uri, verbose = TRUE)
   expect_true(inherits(scdataset, "SCDataset"))
 
@@ -67,6 +59,14 @@ test_that("SCDataset can be created from a Seurat object", {
     expect_identical(embed2[rownames(embed1), ], embed1)
   }
 
+  # check for cell identities
+  # factors are stored in tiledb as character vectors
+  obs_ids <- SeuratObject::Cells(pbmc_small)
+  expect_equal(
+    as.character(SeuratObject::Idents(pbmc_small2)[obs_ids]),
+    as.character(SeuratObject::Idents(pbmc_small)[obs_ids])
+  )
+
   expect_identical(
     SeuratObject::Graphs(pbmc_small2),
     SeuratObject::Graphs(pbmc_small)