Blog - Annotating cells in the pbmc3K dataset

Annotating cells in the pbmc3K dataset (without clustering)

In this blog post, we use CellKb to assign cell types to individual cells in the dataset of Peripheral Blood Mononuclear Cells (PBMC) freely available from 10X Genomics. There are 2,700 single cells that were sequenced on the Illumina NextSeq 500. The raw data can be found here. This is the same dataset analyzed in the Scanpy and Seurat clustering tutorials.

Annotating cells individually is more likely to identify rare cell types. Annotating cells individually will also give a better result when the clusters identified in your dataset are not cleanly separated.

Upload the dataset to CellKb

We upload the dataset to CellKb. CellKb uses Scanpy for reading/pre-processing the dataset. Support for annotation of cells in Seurat objects will be added in the future.

Analyze the dataset

We can now submit the dataset for single-cell annotation, choosing "Blood" in the filter criteria.

Download results

The analysis runs in the background, and you can check the status on the "Results" screen.

After the analysis is complete, you can download the results. The results contain TSV file and PDF files with one plot for each cell type identified, and a summary plot with the broad and detailed cell types identified. We show the summary plot for the pbmc3K dataset here.

Analyzing the pbmc3k dataset using SingleR

We analyze the same pbmc3k dataset using SingleR with the HumanPrimaryCellAtlasData provided by the celldex package as the reference dataset. Our script is as follows.


  library(dplyr)
  library(Seurat) # Seurat version 5.0.1
  library(celldex) # version version 1.12.0
  library(SingleR) # SingleR version 2.4.1
  
  pbmc.data <- Read10X(data.dir="pbmc3k_filtered_gene_bc_matrices\\filtered_gene_bc_matrices\\hg19", gene.column=1)
  pbmc <- CreateSeuratObject(counts=pbmc.data, project="pbmc3k", min.cells=3, min.features=200)
  pbmc[["percent.mt"]] <- PercentageFeatureSet(pbmc, pattern="^MT-")
  pbmc <- NormalizeData(pbmc)
  pbmc <- FindVariableFeatures(pbmc, selection.method="vst", nfeatures=2000)
  all.genes <- rownames(pbmc)
  pbmc <- ScaleData(pbmc, features=all.genes)
  pbmc <- RunPCA(pbmc, features=VariableFeatures(object=pbmc))
  pbmc <- FindNeighbors(pbmc, dims=1:10)
  pbmc <- FindClusters(pbmc, resolution=0.5)
  pbmc <- RunUMAP(pbmc, dims=1:10)
  print(DimPlot(pbmc, reduction="umap"))
  
  ref.data <- celldex::HumanPrimaryCellAtlasData(ensembl=TRUE)
  results_main <- SingleR(test=as.SingleCellExperiment(pbmc), ref=ref.data, labels=ref.data$label.main)
  results_fine <- SingleR(test=as.SingleCellExperiment(pbmc), ref=ref.data, labels=ref.data$label.fine)
  
  pbmc$celltype_broad <- results_main$labels
  pbmc$celltype_detailed <- results_fine$labels
  print(DimPlot(pbmc, reduction="umap", group.by='celltype_broad', repel=T, label=TRUE))
  print(DimPlot(pbmc, reduction="umap", group.by='celltype_detailed', repel=T, label=TRUE))
  
  results_fine$labels_main = results_main$labels
  write.table(results_fine, file="singler_results_without_clustering.txt", sep="\t", row.names=TRUE)
  write.table(table(results_fine$labels), file="singler_results_without_clustering_summary.txt", sep="\t", row.names=TRUE)

We show the summary plot with the broad and detailed cell types identified by SingleR for the pbmc3K dataset here.

Comparing the results (CellKb vs SingleR)

You will notice that while the broad cell types are similar, CellKb has identified more detailed cell types than SingleR.

Category	Name of the cell type in SingleR	Cell count	Name of the cell type in CellKb	Cell count
T cells (CD4+)	T_cell:CD4+_central_memory	610	central memory CD4 positive, alpha-beta T cell	23
T cells (CD4+)	T_cell:CD4+	28	CD4-positive, alpha-beta T cell	408
T cells (CD4+)	T_cell:CD4+_effector_memory	347	effector memory CD4-positive, alpha-beta T cell	156
T cells (CD4+)	T_cell:CD4+_naive	164	naive thymus-derived CD4-positive, alpha-beta T cell	23
T cells (CD4+)	-	-	CD4-positive, alpha-beta memory T cell	119
T cells (CD4+)	-	-	CD4-positive helper T cell	23
T cells (CD4+)	-	-	CD4-positive, alpha-beta cytotoxic T cell	11
T cells (CD4+)	-	-	CD4-positive, CD25-positive, alpha-beta regulatory T cell	4
T cells (CD4+)	-	-	effector CD4-positive, alpha-beta T cell	2
T cells (CD4+)	-	-	T follicular helper cell	2
T cells (CD4+)	-	-	T-helper 1 cell	2
T cells (CD4+)	-	-	T-helper 17 cell	2
T cells (CD8+)	T_cell:CD8+	240	CD8-positive, alpha-beta T cell	94
T cells (CD8+)	T_cell:CD8+_effector_memory	48	effector memory CD8-positive, alpha-beta T cell	62
T cells (CD8+)	T_cell:CD8+_naive	9	naive thymus-derived CD8-positive, alpha-beta T cell	90
T cells (CD8+)	T_cell:CD8+_effector_memory_RA	5	CD8-positive, alpha-beta memory T cell	55
T cells (CD8+)	T_cell:CD8+_central_memory	4	central memory CD8-positive, alpha-beta T cell	30
T cells (CD8+)	-	-	CD8-positive, alpha-beta cytotoxic T cell	4
T cells (CD8+)	-	-	effector CD8-positive, alpha-beta T cell	4
T cells (CD8+)	-	-	activated CD8-positive, alpha-beta T cell	12
T cells (CD8+)	-	-	effector memory CD8-positive, alpha-beta T cell, terminally differentiated	2
T cells (other)	T_cell:gamma-delta	5	gamma-delta T cell	14
T cells (other)	-	-	T cell	102
T cells (other)	-	-	mucosal invariant T cell	82
T cells (other)	-	-	regulatory T cell	51
T cells (other)	-	-	naive T cell	50
T cells (other)	-	-	exhausted T cell	2
T cells (other)	-	-	mature NK T cell	20
T cells (other)	-	-	immature NK T cell	1
T cells (other)	-	-	memory T cell	1
NK cells	NK_cell	141	natural killer cell	162
NK cells	NK_cell:CD56hiCD62L+	31	CD16-negative, CD56-bright natural killer cell, human	10
NK cells	NK_cell:IL2	8	mature NK T cell	20
NK cells	-	-	immature NK T cell	1
Monocytes	Monocyte:CD16-	411	classical monocyte	117
Monocytes	Monocyte:CD16+	259	non-classical monocyte	149
Monocytes	Monocyte	4	monocyte	320
Monocytes	Monocyte:leukotriene_D4	1	CD14-positive monocyte	28
Monocytes	-	-	CD14-low, CD16-positive monocyte	25
Monocytes	-	-	CD14-positive, CD16-positive monocyte	4
Monocytes	-	-	intermediate monocyte	3
B cells	B_cell:immature	216	-
B cells	B_cell	32	B cell	268
B cells	B_cell:Naive	82	naive B cell	39
B cells	B_cell:Memory	19	memory B cell	36
B cells	B_cell:Plasma_cell	4	plasma cell	3
B cells	-	-	class switched memory B cell	1
Dendritic cells	-	-	conventional dendritic cell	21
Dendritic cells	-	-	dendritic cell	19
Dendritic cells	-	-	plasmacytoid dendritic cell	3
Platelets	Platelets	9	platelet	21
Platelets	-	-	megakaryocyte	1
Other	Pre-B_cell_CD34-	17	double negative thymocyte	3
Other	CMP	4	erythrocyte	1
Other	MEP	1	group 1 innate lymphoid cell	1
Other	Neutrophil:commensal_E._coli_MG1655	1	lymphocyte	1
Other	-	-	progenitor cell	1

We also notice that dendritic cells were incorrectly annotated by SingleR/celldex.

Conclusion

CellKb uses a large set of publications as a reference, and is able to identify cell types to a more detailed resolution. CellKb uses standardized ontology terms in the results, making it easier to integrate in your pipelines. In this blog post, we have shown how you can annotate individual cells using CellKb.