Tinder for single-cell data

Overview

This package provides Python bindings to the C++ implementation of the SingleR method, originally developed by Aran et al. (2019). It is designed to annotate cell types by matching cells to known references based on their expression profiles. So kind of like Tinder, but for cells.

Quick start

Firstly, let's load in the famous PBMC 4k dataset from 10X Genomics. Any SummarizedExperiment can be used here.

importsinglecellexperimentsce=singlecellexperiment.read_tenx_h5("pbmc4k-tenx.h5", realize_assays=True)
## class: SingleCellExperiment## dimensions: (33694, 4340)## assays(1): ['counts']## row_data columns(2): ['id', 'name']## row_names(0):## column_data columns(0): []## column_names(0):## main_experiment_name:## reduced_dims(0): []## alternative_experiments(0): []## row_pairs(0): []## column_pairs(0): []## metadata(0):

Now, we fetch the Blueprint/ENCODE reference from the celldex package:

importcelldexref_data=celldex.fetch_reference("blueprint_encode", "2024-02-26", realize_assays=True)
## class: SummarizedExperiment## dimensions: (19859, 259)## assays(1): ['logcounts']## row_data columns(0): []## row_names(19859): ['TSPAN6', 'TNMD', 'DPM1', ..., 'MIR522', 'LINC00550', 'GIMAP1-GIMAP5']## column_data columns(3): ['label.main', 'label.fine', 'label.ont']## column_names(259): ['mature.neutrophil', 'CD14.positive..CD16.negative.classical.monocyte', 'mature.neutrophil.1', ..., 'fibroblast.of.dermis.1', 'epithelial.cell.of.umbilical.artery.1', 'dermis.lymphatic.vessel.endothelial.cell.1']## metadata(0): 

We annotate each cell in sce against the reference. This yields a data frame that contains all of the assignments and the scores for each label:

importsinglerresults=singler.annotate_single(
test_data=sce,
test_features=sce.get_row_data()["name"],
ref_data=ref_data,
ref_labels=ref_data.get_column_data().column("label.main"),
)
print(results)
## BiocFrame with 4340 rows and 3 columns## best scores delta## <list> <BiocFrame> <ndarray[float64]>## [0] Monocytes 0.2562168476981947:0.1254343439610945... 0.4378177347327983## [1] Monocytes 0.2834593285584352:0.1350551446328624... 0.06708042619997218## [2] Monocytes 0.27001789110872965:0.149733483922888... 0.29630159290612557## ... ... ...## [4337] NK cells 0.22504679944584366:0.128832705528845... 0.09253938940916262## [4338] B-cells 0.21466213533061748:0.143717963254983... 0.06727011631382662## [4339] Monocytes 0.2880677943712168:0.1327331541412791... 0.06576621116161818## ------## metadata(2): used markersprint(results["scores"]["Macrophages"])
## [0.3553803 0.40346796 0.3680465 ... 0.32339334 0.29082273 0.39644526]

Calling low-level functions

The annotate_single() function is a convenient wrapper around a number of lower-level functions in singler. Advanced users may prefer to build the reference and run the classification separately. This allows us to re-use the same reference for multiple datasets without repeating the build step.

built=singler.train_single(
ref_data=ref_data,
ref_labels=ref_data.get_column_data().column("label.main"),
ref_features=ref_data.get_row_names(),
test_features=sce.get_row_data()["name"]
)

Then, we apply the pre-built reference to the test dataset to obtain our label assignments. This can be repeated with different datasets that have the same features as test_features=.

output=singler.classify_single(mat, ref_prebuilt=built)
print(output)
## BiocFrame with 4340 rows and 3 columns## best scores delta## <list> <BiocFrame> <ndarray[float64]>## [0] Monocytes 0.2562168476981947:0.1254343439610945... 0.4378177347327983## [1] Monocytes 0.2834593285584352:0.1350551446328624... 0.06708042619997218## [2] Monocytes 0.27001789110872965:0.149733483922888... 0.29630159290612557## ... ... ...## [4337] NK cells 0.22504679944584366:0.128832705528845... 0.09253938940916262## [4338] B-cells 0.21466213533061748:0.143717963254983... 0.06727011631382662## [4339] Monocytes 0.2880677943712168:0.1327331541412791... 0.06576621116161818## ------## metadata(2): used markers

Integrating labels across references

We can use annotations from multiple references through the annotate_integrated() function. This annotates the test dataset against each reference individually to obtain the best per-reference label, and then it compares across references to find the best label from all references.

importsinglerimportcelldexblueprint_ref=celldex.fetch_reference("blueprint_encode", "2024-02-26", realize_assays=True)
immune_cell_ref=celldex.fetch_reference("dice", "2024-02-26", realize_assays=True)
integrated_res=singler.annotate_integrated(
mat,
ref_data= [
blueprint_ref,
immune_cell_ref
],
ref_labels= [
blueprint_ref.get_column_data().column("label.main"),
immune_cell_ref.get_column_data().column("label.main")
],
test_features=features
)
print(integrated_res["integrated"])
## BiocFrame with 4340 rows and 4 columns## best_label best_reference scores delta## <list> <ndarray[uint32]> <BiocFrame> <ndarray[float64]>## [0] Monocytes 0 Monocytes:0.4601040318745395:Monocyte... 0.07172619402931646## [1] Monocytes 0 Monocytes:0.5569436588644365:Monocyte... 0.10337145230321299## [2] Monocytes 0 Monocytes:0.460675384672641:Monocytes... 0.06302300967458618## ... ... ... ...## [4337] NK cells 0 NK cells:0.5639386082584756:NK cells:... 0.02453897370863012## [4338] B-cells 0 B-cells:0.49462921210156113:B cells:0... 0.0259893105975339## [4339] Monocytes 0 Monocytes:0.49997247809330014:Monocyt... 0.08744894116986357

The best_label columns contains the best label for each cell across all references, while the best_reference specifies the reference (in the same order as ref_data) that contains the best label.