
Automatic annotation of cell types
AnnotateCells.RdThis function transfers cell type annotations from a reference Seurat dataset
to a query Seurat dataset using non-negative least squares ("nnls", default)
or Seurat anchor transfer ("Seurat"). For each column specified in
reference_groups, a corresponding column of predicted cell type annotations
and a <group>_score column of per-cell prediction scores are added to the
metadata of the query dataset. In addition, one can optionally
summarize these annotations for a specific clustering solution, which adds a
<group>_summary column containing the most common annotation per cluster.
Usage
AnnotateCells(
object,
reference = reference,
summarize_by_column = NULL,
reference_assay = "PNA",
query_assay = "PNA",
reference_groups = c("celltype.l1", "celltype.l2"),
normalization_method = c("CLR", "LogNormalize", "SCT"),
reduction = "cca",
method = c("nnls", "Seurat", "nmf"),
min_prediction_score = 0,
predict_doublets = FALSE,
min_doublet_score = 0.4,
skip_normalization = FALSE,
verbose = TRUE,
...
)Arguments
- object
Seurat object to which you want to add celltype annotation
- reference
Seurat object that contains annotated reference data, see
read_pbmc_referenceand examples for obtaining a PBMC reference dataset- summarize_by_column
If NULL, return only the cell type annotation per cell. If the name of a column in the object metadata is provided, the annotations will be summarized to the factors of this column, i.e. the most common annotation per factor level will be retained. Useful for annotating the output of the Seurat clustering, for example, Default: NULL
- reference_assay
Assay in the reference dataset used for annotation, Default: 'PNA'
- query_assay
Assay in the query dataset used for annotation, Default: 'PNA'
- reference_groups
A character vector of reference metadata columns to use for annotation, Default:
c("celltype.l1", "celltype.l2")- normalization_method
Normalization method used prior to annotation (with margin = 2), either "CLR", "LogNormalize", or "SCT", Default: 'CLR'
- reduction
Reduction method used during
FindTransferAnchorswhenmethod = "Seurat", Default: 'cca'- method
Method to use for annotation, either "nnls", "Seurat", or "nmf" (alias for "nnls"), Default: 'nnls'
- min_prediction_score
Minimum prediction score for the annotation to be considered valid. Labels with a prediction score below this value will be labelled as "Unknown", Default: 0, meaning that no threshold is used.
- predict_doublets
If TRUE, compute a heterotypic doublet score and candidate cell type pair using the first group in
reference_groupswhenmethod = "nnls", addingdoublet_scoreanddoublet_paircolumns to the object metadata, Default: FALSE- min_doublet_score
Minimum doublet score required to report a candidate cell type pair in
doublet_pairwhenpredict_doublets = TRUE. Cells with a doublet score below this threshold will havedoublet_pairset toNA, Default: 0.4- skip_normalization
If TRUE, the datasets will not be normalized prior to annotation. The default layer will be used, so use this setting at your own risk, Default: FALSE
- verbose
If TRUE, print messages about the progress of the annotation, Default: TRUE
- ...
Additional parameters passed to
FindTransferAnchorswhenmethod = "Seurat"
Details
The "nnls" (or "nmf") method projects query cells onto feature-weighted reference
cell type profiles using non-negative least squares (RcppML::project), followed by
local fine-tuning across the top candidate cell types. The "Seurat" method is a wrapper for
the FindTransferAnchors and TransferData functions from Seurat. Both methods
support an optional summary per cluster.
Merged SeuratObject v5 assays (Assay5, including PNAAssay5) must not
retain split merge layers (counts.1, data.1, etc.).
Run JoinLayers() on the Seurat object (or on the assay) after merge()
before calling AnnotateCells().
Examples
if (FALSE) { # \dontrun{
reference <- read_pbmc_reference()
seur <- ReadPNA_Seurat(minimal_pna_pxl_file())
seur <- AnnotateCells(
seur,
reference = reference
)
} # }