单细胞 | label transfer with Seurat4(未知细胞映射到注释好的细胞图谱)

场景：把新的细胞比对到已经注释过的细胞集合上，获取映射后的细胞标签，UMP坐标。
准备：

• 一个分析好的单细胞图谱数据集，作为reference数据集。
• 一个新的单细胞counts矩阵，记为 query数据集。

主要分为两个步骤：1.获取分类标签，2.获取UMAP坐标。

1. R语言代码

library(Seurat)

#1. load data ====
pbmc=readRDS("~/data/scScripts/backup/data/pbmc3k_final.rds")
DimPlot(pbmc, label=T)


# 获取细胞子集：CD3很高的细胞
FeaturePlot(pbmc, features = c("CD3D", "CD3E", "CD3G"))
VlnPlot(pbmc, features = c("CD3D", "CD3E", "CD3G"), pt.size = 0)
small0=subset(pbmc, CD3D>3)
pbmc #2638 #as refer, 已知细胞图谱
small0 #228 #as query, 未知细胞

# 未知数据，没有标签
small=CreateSeuratObject(small0@assays$RNA@counts)
#标准化
small=NormalizeData(small)



#2. find anchor ----
anchors <- FindTransferAnchors(reference = pbmc, query = small,
                               dims = 1:10, reference.reduction = "pca")
# Retained 1070 anchors

# get predicted ID
predictions <- TransferData(anchorset = anchors, refdata = pbmc$seurat_clusters,
                            dims = 1:10)
# add metadata
small <- AddMetaData(small, metadata = predictions)
small
table(small$predicted.id)
#  0   1   4 
#104  58  66 
#DimPlot(small, label=T)




#3. UMAP projection----
pbmc2 <- RunUMAP(pbmc, dims = 1:10, reduction = "pca", return.model = TRUE)
pbmc2$celltype=Idents(pbmc2)

DimPlot(pbmc2, label=T)
small <- MapQuery(anchorset = anchors, reference = pbmc2, query = small,
                       #refdata = list(seurat_clusters = "seurat_clusters"),  #可以映射多个标签
                       refdata = list(seurat_clusters = "seurat_clusters", celltype="celltype"), 
                       reference.reduction = "pca", 
                       reduction.model = "umap")
# MapQuery() is a wrapper around three functions: TransferData(), IntegrateEmbeddings(), and ProjectUMAP().

p1 <- DimPlot(pbmc2, reduction = "umap", group.by = "seurat_clusters", label = TRUE, label.size = 3,
              repel = TRUE) + NoLegend() + ggtitle("Reference")
p2 <- DimPlot(small, reduction = "ref.umap", group.by = "predicted.id", label = TRUE,
              label.size = 3, repel = TRUE) + NoLegend() + ggtitle("Query of small")+
  xlim(-7,12)+ ylim(-7,12)
p1 + p2

q1 <- DimPlot(pbmc2, reduction = "umap", group.by = "celltype", label = TRUE, label.size = 3,
              repel = TRUE) + NoLegend() + ggtitle("Reference")
q2 <- DimPlot(small, reduction = "ref.umap", group.by = "predicted.celltype", label = TRUE,
              label.size = 3, repel = TRUE) + NoLegend() + ggtitle("Query of small")+
  xlim(-7,12)+ ylim(-7,12)
q1+q2

#结果的一致性比较
rs=table(small0$seurat_clusters, small$predicted.id)
rs[which(rowSums(rs)!=0),]
#   0  1  4
#0 97  1  0
#1  5 56  0
#4  2  1 64
#6  0  0  2
# 左侧是原始标签，顶部是预测的标签

2. 效果图

图2：

这样，就把未知细胞映射到已经注释好的细胞图谱中了。
除了映射 seurat_cluster 外，还能映射 celltype等自定义metadata。

refer

• https://satijalab.org/seurat/articles/multimodal_reference_mapping.html
• https://satijalab.org/seurat/articles/covid_sctmapping