Annotation of Single Cell Clusters

Question

[Shelle]

I am new to single cell. And I would appreciate guidance for my question

I want to cluster a given single cell dataset for example PBMC. How can I annotate different clusters in the single cell without the knowledge of biology? I came across the scanpy but it was not useful and sounded like you need to know the different type of marker genes beforehand.

Any help would be appreciated

Answer 1

[e.mohammadi.as]

You can use SingleR R package as well. Here is an example:

# pbmc is my Seurat object
# wpath is my working directory

library(Seurat)
library(SingleR)
library(celldex)

## Transform Seurat object to SingleCellExperiment object
Idents(HE_control_2) <- HE_control_2$seurat_clusters
sce.HE_control_2 <- as.SingleCellExperiment(HE_control_2)

wpath="D:/Poland/PHD/spatial/Second_set/Genevariable3000_analysis/avg/correlation/MT15/Final_7000"
## Load references (after loading I save them to RDS file, so next time they load faster)
blue <- SingleR::BlueprintEncodeData()
saveRDS(blue, paste(wpath,"singleR.BlueprintEncodeData.rds",sep="/"))
blue <- readRDS(paste(wpath,"singleR.BlueprintEncodeData.rds",sep="/"))

hpca <- SingleR::HumanPrimaryCellAtlasData()
saveRDS(hpca, paste(wpath,"singleR.HumanPrimaryCellAtlasData.rds",sep="/"))
hpca <- readRDS(paste(wpath,"singleR.HumanPrimaryCellAtlasData.rds",sep="/"))

monaco <- SingleR::MonacoImmuneData()
saveRDS(monaco, paste(wpath,"singleR.MonacoImmuneData.rds",sep="/"))
monaco <- readRDS(paste(wpath,"singleR.MonacoImmuneData.rds",sep="/"))

dice <- SingleR::DatabaseImmuneCellExpressionData()
saveRDS(dice, paste(wpath,"singleR.DatabaseImmuneCellExpressionData.rds",sep="/"))
dice <- readRDS(paste(wpath,"singleR.DatabaseImmuneCellExpressionData.rds",sep="/"))

MTG <- readRDS(paste(wpath,"middle_temporal_gyrus_human_MTG_coarse-grained_annotation.rds",sep="/"))
MTG_fine <- readRDS(paste(wpath,"middle_temporal_gyrus_human_MTG_fine-grained_annotation.rds",sep="/"))

cerebral_cortex_Po <- readRDS(paste(wpath,"cerebral_cortex_human_pollen.rds",sep="/"))
cerebral_cortex_Li <- readRDS(paste(wpath,"colorectal_tumor_hu",sep="/"))



## Check what kind of annotations are available
table(blue$label.main)
table(blue$label.fine)
table(hpca$label.main)
table(hpca$label.fine)
table(MTG$feature_matrix_learned)
table(MTG_fine$cell_type_information)


refs <- list(BP=blue, HPCA=hpca)
fine.labels <- list(blue$label.fine, hpca$label.fine)
main.labels <- list(blue$label.main, hpca$label.main)

## Annotate CELLS (main label)
pred.list <- SingleR(test=sce.HE_control_2, ref=refs, assay.type.test=1, labels=main.labels)
pred.blue <- SingleR(test=sce.pbmc, ref=blue, assay.type.test=1, labels=blue$label.main)
pred.hesc <- SingleR(test=sce.pbmc, ref = hpca, assay.type.test=1, labels=hpca$label.main)
pred.monaco <- SingleR(test=sce.pbmc, ref=monaco, assay.type.test=1, labels=monaco$label.main)
pred.dice <- SingleR(test=sce.pbmc, ref=dice, assay.type.test=1, labels=dice$label.main)

## Annotate CELLS (fine-grained label)
pred.list.fine <- SingleR(test=sce.HE_control_2, ref=refs, assay.type.test=1, labels=fine.labels)
pred.blue2 <- SingleR(test=sce.HE_sample_2, ref=blue, assay.type.test=1, labels=blue$label.fine)
pred.hesc2 <- SingleR(test=sce.pbmc, ref = hpca, assay.type.test=1, labels=hpca$label.fine)
pred.monaco2 <- SingleR(test=sce.pbmc, ref=monaco, assay.type.test=1, labels=monaco$label.fine)
pred.dice2 <- SingleR(test=sce.pbmc, ref=dice, assay.type.test=1, labels=dice$label.fine)
pred.MTG <- SingleR(test=sce.HE_control_2, ref=MTG, assay.type.test=1, labels=MTG$cell_type_information)

## Add annotations to Seurat object
#celllabels1 <- cbind(celllabels=pred.blue$labels, celllabels2=pred.hesc$labels, celllabels3=pred.monaco$labels, celllabels4=pred.dice$labels)
#celllabels2 <- cbind(celllabelsb=pred.blue2$labels, celllabels2b=pred.hesc2$labels, celllabels3b=pred.monaco2$labels, celllabels4b=pred.dice2$labels)
celllabels2 <- cbind(celllabelsb=pred.list$labels)
celllabels1 <- cbind(celllabels=pred.list.fine$labels)

rownames(celllabels1) <- rownames(celllabels2) <- colnames(HE_control_2)
HE_control_2 <- AddMetaData(HE_control_2, as.data.frame(cbind(celllabels1,celllabels2)))

table(pbmc$celllabels, pbmc$seurat_clusters)
table(pbmc$celllabels3, pbmc$seurat_clusters)

## Evaluate annotation based on scores
plotScoreHeatmap(pred.list.fine, clusters=HE_control_2$seurat_clusters)

par(mfrow=c(1,1), mar=c(2.5,7,1,.5))
graphics::boxplot(pred.monaco$scores, xlab="prediction.score", ylab="", main="Prediction scores for cell labels", boxwex=.8,
                  cex=.5, cex.main=.8, cex.lab=.8, cex.axis=.6, las=1, horizontal=T, varwidth=T, col=rainbow(36))

[andynkili]

Do you know if there is a way of using singleR with a list a marker genes (one produced by the findAllMarkers function from Seurat) as a reference?

Answer 2

[GouthamAtla]

One need to know cell-type marker genes to annotate the cell-types. As far as I know, no tool can do it without providing marker genes for cell-type.

[Shelle]

Is there any list of universal marker genes that I can use? For example for 7 clusters in one dataset how I can know which one is representing which cell type marker genes?

[ATpoint]

You have to provide markers based on the biology. Biology and analysis go hand-in-hand. If you have PBMCs then you can get markers from consortium datasets like Haemopedia or ImmGen. Check the singleR package which includes these datasets for hematopoietic cells and can assign identities to your clusters.

[Shelle]

Thanks for the information. Is there any other tool besides singleR which is not really dependent on Seurat? both of these tools are based on R and I would like to stay close to python because of downstream analysis

Answer 3

[Ashwini Patil]

Take a look at the CellKb database. You can enter a ranked gene list and get matching cell types previously published in literature through the web interface. CellKb contains a large number of cell types from 10 species. Academic users can sign up for a free license to search hematopoietic cell types through CellKb Immune at https://www.cellkb.com/immune.

Answer 4

[e.mohammadi.as]

Please how a look at the scCATCH(https://github.com/ZJUFanLab/scCATCH) and SCSA (https://github.com/bioinfo-ibms-pumc/SCSA).

Answer 5

[Jennifer Pham]

Hi @Shelle, you may try our tool for automatic cell type annotation here: talk2data.bioturing.com/predict. It covers quite diverse cell types and subtypes.