拟时序涉及到取亚群，这里我在文献里看到过，例如，取成纤维细胞是选择 PDGFRΑ 基因阳性的细胞。但是PBMC的单核细胞好像不是用某个基因取的子集。所以，这个应该是不同细胞亚群，有不同的取子集方法。

一、按照基因阳性取

PDGFRΑ⁺ stromal fibroblasts from the colon were distinguishable in multiple cell clusters in all mice (Figure 3A).Skin inflammation activates intestinal stromal fibroblasts and promotes colitis - PMC (nih.gov)

UMAP plot after subclustering of Pdgfra+ fibroblasts, showing 10 distinct subtypes (0 to 9) colored by cluster. Antimicrobial production by perifollicular dermal preadipocytes is essential to the pathophysiology of acne - PMC (nih.gov)

转自：玩转单细胞---提取特定基因表达的细胞群分析（一个小问题

读之前帖子的文献，发现一个问题，后面单细胞亚群分析的时候它是提取的表达特定基因的细胞群，而不是直接提取命名好的细胞群。虽然只是一个小问题，还是值得说一下，这也是一种思路，有时候可能仅关注特定表达基因细胞群分析会更加有意义！
这是已经定好群的细胞：

library(Seurat)
library(dplyr)
DimPlot(mouse_data, label = T)+NoLegend()
FeaturePlot(mouse_data, features = 'Il1b', pt.size = 1)

image.png

可以看出，Il1b的表达并不是在特定细胞群，而是有穿插，我们需要研究这群细胞，就需要将其提取。思路很简单，那就是获得表达Il1b基因细胞的ID，然后提取即可：

expr <- mouse_data@assays$RNA@data
gene_expression <- expr %>% .["Il1b",] %>% as.data.frame()
colnames(gene_expression) <- "Il1b"
gene_expression$cell <- rownames(gene_expression)

#选择阳性细胞
gene_expression_sel <- gene_expression[which(gene_expression$Il1b>0),]
mouse_sel_sce <- mouse_data[,rownames(gene_expression_sel)]

作图看看：

DimPlot(mouse_sel_sce, label = T, pt.size = 1)+NoLegend()
FeaturePlot(mouse_sel_sce, features = 'Il1b', pt.size = 1)

image.png

可以看出，这就是我们需要选择的细胞！之后亚群分析走标准流程即可！虽然是一个小问题，但还是挺有用的，觉得分享有用的点个赞、关注一下呗！

二、就按照普通方法取

scRNA-Seq | 单细胞亚群合并与提取 - 简书 (jianshu.com)

levels(sce)
# [1] "Naive CD4 T"  "CD14+ Mono"   "Memory CD4 T" "B"            "CD8 T"        "FCGR3A+ Mono"
# [7] "NK"           "DC"           "Platelet" 

genes_to_check = c('PTPRC', 'CD3D', 'CD3E', 
                   'CD4','IL7R','NKG7','CD8A')

p1=DimPlot(sce, reduction = 'umap', group.by = 'seurat_clusters',
           label = TRUE, pt.size = 0.5) + NoLegend()
p2=DotPlot(sce, group.by = 'seurat_clusters',
           features = unique(genes_to_check)) + RotatedAxis()
p1+p2

image.png

假设这个时候，我们想提取CD4的T细胞，那么根据上文聚类0/2/4/6均为T细胞，其中0和2表达CD4相对4/6较高，但是其实示例里面的CD4的T细胞并不怎么高表达CD4，在此不深究，继续向下走。

提取指定单细胞亚群，这三种取法都可以

cd4_sce1 = sce[,sce@meta.data$seurat_clusters %in% c(0,2)]
cd4_sce2 = sce[, Idents(sce) %in% c( "Naive CD4 T" ,  "Memory CD4 T" )]
cd4_sce3 = subset(sce,seurat_clusters %in% c(0,2))
## 较简单，核心原理就是R里取子集的3种策略：逻辑值，坐标，名字

重新降维聚类分群

sce=cd4_sce1
sce <- NormalizeData(sce, normalization.method = "LogNormalize", scale.factor = 1e4) 
sce <- FindVariableFeatures(sce, selection.method = 'vst', nfeatures = 2000)
sce <- ScaleData(sce, vars.to.regress = "percent.mt")
sce <- RunPCA(sce, features = VariableFeatures(object = sce)) 
sce <- FindNeighbors(sce, dims = 1:10)
sce <- FindClusters(sce, resolution = 1 )
head(Idents(sce), 5)
table(sce$seurat_clusters) 
sce <- RunUMAP(sce, dims = 1:10)
DimPlot(sce, reduction = 'umap')

genes_to_check = c('PTPRC', 'CD3D', 'CD3E', 'FOXP3',
                   'CD4','IL7R','NKG7','CD8A')
DotPlot(sce, group.by = 'seurat_clusters',
        features = unique(genes_to_check)) + RotatedAxis()
# 亚群水平 
p1=DimPlot(sce, reduction = 'umap', group.by = 'seurat_clusters',
           label = TRUE, pt.size = 0.5) + NoLegend()
p2=DotPlot(sce, group.by = 'seurat_clusters',
           features = unique(genes_to_check)) + RotatedAxis()
p1+p2

取子集 - 简书 (jianshu.com)

R里面取子集可以根据，逻辑值 或者 是坐标 取。
逻辑值 是判断要不要取，坐标是告诉他取哪个坐标

cd4_sce1 = sec [,sce@metadata$seurat_clusters %in% c(0,2)]  # 逻辑值
cd4_sce2 = sec [, Idents(sce) %in% c("Naive CD4T","Memory CD4T")]  # 和上面的效果一样
# subset 函数也可以