原文链接Chapter 12 Cell type annotation

1、概述

• straightforward annotation approach is to compare the single-cell expression profiles with previously annotated reference datasets.

• 其中最关键的就是reference datasets参考数据
  关于参考数据，本质上就是sce对象，其中colData slot 含有cell type 的 label信息

• 本文笔记主要基于SingleR包的注释方法，而且该包也内置了许多 reference data可供使用。
  

  SingleR 内置数据集概况

2、SingleR注释

（1）基本方法

#加载待注释sce
load("fluidigm.clust.RData")
fluidigm.clust

#准备合适的ref data
library(SingleR)
ref <- BlueprintEncodeData()
ref

pred <- SingleR(test=fluidigm.clust, ref=ref, labels=ref$label.main)
#pred <- SingleR(test=fluidigm.clust, ref=ref, labels=ref$label.fine)
table(pred$labels)

• ref$label.fine provides more resolution at the cost of speed and increased ambiguity in the assignments.
  简单来说就是reflabel.fine分得细

2-1

fluidigm.clust
colnames(colData(fluidigm.clust))
fluidigm.clust$celltype <- pred$labels
table(fluidigm.clust$celltype)
plotReducedDim(fluidigm.clust, dimred="UMAP", colour_by="celltype")
fluidigm.anno <- fluidigm.clust
save(fluidigm.anno,file = "fluidigm.anno.Rdata")

（2）visualization digonosis

• heatmap
  每一列为细胞与细胞类型（行）的比对情况，列标注取比对值最高对应的细胞类型

plotScoreHeatmap(pred)

plotScoreHeatmap(pred)

• jitter and violin plots
  showing assignment scores or related values for all cells across one or more labels.

sum(is.na(pred$pruned.labels)) 
#无 pruned cell
plotScoreDistribution(pred)
#black point for each cell
#grey area for cells that were assigned to the label.
#yellow area for other cells not assigned to the label.

plotScoreDistribution(pred)

• 最后还可以比较下已知注释分类与singler预测分类的关系

tab <- table(Assigned=pred$pruned.labels, Cluster=fluidigm.clust$Cluster2)
tab
# Adding a pseudo-count of 10 to avoid strong color jumps with just 1 cell.
library(pheatmap)
pheatmap(log2(tab+10), color=colorRampPalette(c("white", "blue"))(101))

ref data from other source

• 代表性的就是scRNAseq contains many single-cell datasets, many of which contain the authors’ manual annotations.可以用来当做ref data。

library(scRNAseq)
sceM <- MuraroPancreasData()
sceM
#此外要注意的是基因名为Ensemble ID
table(sceM$label)

sceM

• 待分类数据

#ID转换：symbol→ensemble
library(AnnotationHub)
edb <- AnnotationHub()[["AH73881"]]
gene.symb <- sub("__chr.*$", "", rownames(sceG))
gene.ids <- mapIds(edb, keys=gene.symb, 
                   keytype="SYMBOL", column="GENEID")
keep <- !is.na(gene.ids) & !duplicated(gene.ids)
sceG <- sceG[keep,]
rownames(sceG) <- gene.ids[keep]
counts(sceG)[1:4,1:4]
sceG

• 注释

pred.sceG <- SingleR(test=sceG, ref=sceM, 
                      labels=sceM$label, de.method="wilcox")
table(pred.sceG$labels)

3、其它注释方法

简单介绍，不再操作，详见原文

（1）Assigning cell labels from gene sets

• A related strategy is to explicitly identify sets of marker genes that are highly expressed in each individual cell.
• 简单来说是比较特定细胞代表基因特征与待分类sce的每一个细胞的表达概况的相似度，以AUC曲线为指标确定最符合的cell type

（2）Assigning cluster labels from markers

• Yet another strategy for annotation is to perform a gene set enrichment analysis on the marker genes defining each cluster.
• This identifies the pathways and processes that are (relatively) active in each cluster based on upregulation of the associated genes compared to other clusters.
• 简单来说，就是对每个clust的marker基因进行go/kegg点的富集分析，通过对应结果的discription确定cell type

---

以上是第十二章Clustering部分的简单流程笔记，主要学习了基于SingleR的cell type注释方法。其它方式详见原文Chapter 12 Cell type annotation
本系列笔记基于OSCA全流程的大致流程梳理，详细原理可参考原文。如有错误，恳请指正！
此外还有刘小泽老师整理的全文翻译笔记，详见目录。