一、总体代码

Seurat包官方文档https://satijalab.org/seurat/articles/pbmc3k_tutorial


#加载所需R包
library(Seurat)
library(tidyverse)
library(dplyr)
library(patchwork)
library(ggplot2)
#读取10X数据
scRNA.counts=Read10X(“~/path)
#创建Seurat对象
scRNA <- CreateSeuratObject(scRNA.counts ,min.cells = 3,project="os", min.features = 300)
#去除线粒体和红细胞相关的基因
##去除线粒体
scRNA[["percent.mt"]] <- PercentageFeatureSet(scRNA, pattern = "^MT-")
##红细胞基因比例
HB.genes <- c("HBA1","HBA2","HBB","HBD","HBE1","HBG1","HBG2","HBM","HBQ1","HBZ")
HB_matched <- match(HB.genes, rownames(scRNA@assays$RNA)) 
HB.genes <- rownames(scRNA@assays$RNA)[HB_matched] 
HB.genes <- HB.genes[!is.na(HB.genes)] 
scRNA[["percent.HB"]]<-PercentageFeatureSet(scRNA, features=HB.genes) 
#观察数据结构
levels(scRNA@active.ident)
col.num <- length(levels(scRNA@active.ident))
violin <- VlnPlot(scRNA,
                  features = c("nFeature_RNA", "nCount_RNA", "percent.mt","percent.HB"), 
                  cols =rainbow(col.num), 
                  pt.size = 0.01, 
                  ncol = 4) + 
  theme(axis.title.x=element_blank(), axis.text.x=element_blank(), axis.ticks.x=element_blank()) 
#均一化数据
scRNAa <- NormalizeData(scRNAa, normalization.method = "LogNormalize", scale.factor = 10000)
#寻找高变基因
scRNAa <- FindVariableFeatures(scRNAa, selection.method = "vst", nfeatures = 2000) 
top10 <- head(VariableFeatures(scRNA1), 10) 
plot1 <- VariableFeaturePlot(scRNA1) 
plot2 <- LabelPoints(plot = plot1, points = top10, repel = TRUE, size=2.5) 
plot <- CombinePlots(plots = list(plot1, plot2),legend="bottom") 
#数据缩放
scale.genes <-  rownames(scRNAa)
scRNAb <- ScaleData(scRNAa, features = scale.genes)
#去除周期的影响
CaseMatch(c(cc.genes$s.genes,cc.genes$g2m.genes),VariableFeatures(scRNA1))
g2m_genes = cc.genes$g2m.genes
g2m_genes = CaseMatch(search = g2m_genes, match = rownames(scRNA1))
s_genes = cc.genes$s.genes
s_genes = CaseMatch(search = s_genes, match = rownames(scRNA1))
scRNA11 <- CellCycleScoring(object=scRNA11,  g2m.features=g2m_genes,  s.features=s_genes)
scRNA2 <- CellCycleScoring(object=scRNA2,  g2m.features=g2m_genes,  s.features=s_genes)
#去除周期的影响
library(future)
plan("multicore",workers = 128)
plan("multisession", workers = 64)
options(future.globals.maxSize= 2147483648)
scRNAb <- ScaleData(scRNA11, vars.to.regress = c("S.Score", "G2M.Score"), features = rownames(scRNA1))
#数据降维
scRNAb1 <- RunPCA(scRNAb, features = VariableFeatures(scRNAb)) 
plot1 <- DimPlot(scRNAb1, reduction = "pca", group.by="orig.ident") 
plot1
ElbowPlot(scRNAb1, ndims=20, reduction="pca") 
pc.num=1:20
scRNAb1 <- FindNeighbors(scRNAb1, dims = pc.num) 
scRNAb1 <- FindClusters(scRNAb1, resolution = 1.0)
scRNAb1<-BuildClusterTree(scRNAb1)
install.packages('ape')
PlotClusterTree(scRNAb1)
scRNAb1 = RunTSNE(scRNAb1, dims = pc.num)
embed_tsne <- Embeddings(scRNAb1, 'tsne')
write.csv(embed_tsne,'embed_tsne.csv')
plot1 = DimPlot(scRNAb1, reduction = "tsne") 
plot1
DimPlot(scRNAb1, reduction = "tsne",label = TRUE) 
ggsave("tSNE.pdf", plot = plot1, width = 8, height = 7)
scRNAb1 <- RunUMAP(scRNAb1, dims = pc.num)
embed_umap <- Embeddings(scRNAb1, 'umap')
write.csv(embed_umap,'embed_umap.csv') 
plot2 = DimPlot(scRNAb1, reduction = "umap") 
plot2
#加载所需R包
library(Seurat)
library(tidyverse)
library(dplyr)
library(patchwork)
library(ggplot2)

#读取10X数据
scRNA.counts=Read10X(“~/path)

#创建Seurat对象
scRNA <- CreateSeuratObject(scRNA.counts ,min.cells = 3,project="os", min.features = 300)

#去除线粒体和红细胞相关的基因
##去除线粒体
scRNA[["percent.mt"]] <- PercentageFeatureSet(scRNA, pattern = "^MT-")
##红细胞基因比例
HB.genes <- c("HBA1","HBA2","HBB","HBD","HBE1","HBG1","HBG2","HBM","HBQ1","HBZ")
HB_matched <- match(HB.genes, rownames(scRNA@assays$RNA)) 
HB.genes <- rownames(scRNA@assays$RNA)[HB_matched] 
HB.genes <- HB.genes[!is.na(HB.genes)] 
scRNA[["percent.HB"]]<-PercentageFeatureSet(scRNA, features=HB.genes) 

#观察数据结构
levels(scRNA@active.ident)
col.num <- length(levels(scRNA@active.ident))
violin <- VlnPlot(scRNA,
                  features = c("nFeature_RNA", "nCount_RNA", "percent.mt","percent.HB"), 
                  cols =rainbow(col.num), 
                  pt.size = 0.01, 
                  ncol = 4) + 
  theme(axis.title.x=element_blank(), axis.text.x=element_blank(), axis.ticks.x=element_blank()) 

#均一化数据
scRNAa <- NormalizeData(scRNAa, normalization.method = "LogNormalize", scale.factor = 10000)

#寻找高变基因
scRNAa <- FindVariableFeatures(scRNAa, selection.method = "vst", nfeatures = 2000) 
top10 <- head(VariableFeatures(scRNA1), 10) 

plot1 <- VariableFeaturePlot(scRNA1) 

plot2 <- LabelPoints(plot = plot1, points = top10, repel = TRUE, size=2.5) 
plot <- CombinePlots(plots = list(plot1, plot2),legend="bottom") 

#数据缩放
scale.genes <-  rownames(scRNAa)
scRNAb <- ScaleData(scRNAa, features = scale.genes)

#去除周期的影响
CaseMatch(c(cc.genes$s.genes,cc.genes$g2m.genes),VariableFeatures(scRNA1))
g2m_genes = cc.genes$g2m.genes
g2m_genes = CaseMatch(search = g2m_genes, match = rownames(scRNA1))
s_genes = cc.genes$s.genes
s_genes = CaseMatch(search = s_genes, match = rownames(scRNA1))
scRNA11 <- CellCycleScoring(object=scRNA11,  g2m.features=g2m_genes,  s.features=s_genes)
scRNA2 <- CellCycleScoring(object=scRNA2,  g2m.features=g2m_genes,  s.features=s_genes)

#去除周期的影响
library(future)
plan("multicore",workers = 128)
plan("multisession", workers = 64)
options(future.globals.maxSize= 2147483648)

scRNAb <- ScaleData(scRNA11, vars.to.regress = c("S.Score", "G2M.Score"), features = rownames(scRNA1))

#数据降维

scRNAb1 <- RunPCA(scRNAb, features = VariableFeatures(scRNAb)) 
plot1 <- DimPlot(scRNAb1, reduction = "pca", group.by="orig.ident") 
plot1

ElbowPlot(scRNAb1, ndims=20, reduction="pca") 
pc.num=1:20
scRNAb1 <- FindNeighbors(scRNAb1, dims = pc.num) 
scRNAb1 <- FindClusters(scRNAb1, resolution = 1.0)

scRNAb1<-BuildClusterTree(scRNAb1)
install.packages('ape')
PlotClusterTree(scRNAb1)

scRNAb1 = RunTSNE(scRNAb1, dims = pc.num)
embed_tsne <- Embeddings(scRNAb1, 'tsne')
write.csv(embed_tsne,'embed_tsne.csv')
plot1 = DimPlot(scRNAb1, reduction = "tsne") 

plot1

DimPlot(scRNAb1, reduction = "tsne",label = TRUE) 

ggsave("tSNE.pdf", plot = plot1, width = 8, height = 7)

scRNAb1 <- RunUMAP(scRNAb1, dims = pc.num)
embed_umap <- Embeddings(scRNAb1, 'umap')
write.csv(embed_umap,'embed_umap.csv') 
plot2 = DimPlot(scRNAb1, reduction = "umap") 
plot2
#加载所需R包
library(Seurat)
library(tidyverse)
library(dplyr)
library(patchwork)
library(ggplot2)

#读取10X数据
scRNA.counts=Read10X(“~/path)

#创建Seurat对象
scRNA <- CreateSeuratObject(scRNA.counts ,min.cells = 3,project="os", min.features = 300)

#去除线粒体和红细胞相关的基因
##去除线粒体
scRNA[["percent.mt"]] <- PercentageFeatureSet(scRNA, pattern = "^MT-")
##红细胞基因比例
HB.genes <- c("HBA1","HBA2","HBB","HBD","HBE1","HBG1","HBG2","HBM","HBQ1","HBZ")
HB_matched <- match(HB.genes, rownames(scRNA@assays$RNA)) 
HB.genes <- rownames(scRNA@assays$RNA)[HB_matched] 
HB.genes <- HB.genes[!is.na(HB.genes)] 
scRNA[["percent.HB"]]<-PercentageFeatureSet(scRNA, features=HB.genes) 

#观察数据结构
levels(scRNA@active.ident)
col.num <- length(levels(scRNA@active.ident))
violin <- VlnPlot(scRNA,
                  features = c("nFeature_RNA", "nCount_RNA", "percent.mt","percent.HB"), 
                  cols =rainbow(col.num), 
                  pt.size = 0.01, 
                  ncol = 4) + 
  theme(axis.title.x=element_blank(), axis.text.x=element_blank(), axis.ticks.x=element_blank()) 

#均一化数据
scRNAa <- NormalizeData(scRNAa, normalization.method = "LogNormalize", scale.factor = 10000)

#寻找高变基因
scRNAa <- FindVariableFeatures(scRNAa, selection.method = "vst", nfeatures = 2000) 
top10 <- head(VariableFeatures(scRNA1), 10) 

plot1 <- VariableFeaturePlot(scRNA1) 

plot2 <- LabelPoints(plot = plot1, points = top10, repel = TRUE, size=2.5) 
plot <- CombinePlots(plots = list(plot1, plot2),legend="bottom") 

#数据缩放
scale.genes <-  rownames(scRNAa)
scRNAb <- ScaleData(scRNAa, features = scale.genes)

#去除周期的影响
CaseMatch(c(cc.genes$s.genes,cc.genes$g2m.genes),VariableFeatures(scRNA1))
g2m_genes = cc.genes$g2m.genes
g2m_genes = CaseMatch(search = g2m_genes, match = rownames(scRNA1))
s_genes = cc.genes$s.genes
s_genes = CaseMatch(search = s_genes, match = rownames(scRNA1))
scRNA11 <- CellCycleScoring(object=scRNA11,  g2m.features=g2m_genes,  s.features=s_genes)
scRNA2 <- CellCycleScoring(object=scRNA2,  g2m.features=g2m_genes,  s.features=s_genes)

#去除周期的影响
library(future)
plan("multicore",workers = 128)
plan("multisession", workers = 64)
options(future.globals.maxSize= 2147483648)

scRNAb <- ScaleData(scRNA11, vars.to.regress = c("S.Score", "G2M.Score"), features = rownames(scRNA1))

#数据降维

scRNAb1 <- RunPCA(scRNAb, features = VariableFeatures(scRNAb)) 
plot1 <- DimPlot(scRNAb1, reduction = "pca", group.by="orig.ident") 
plot1

ElbowPlot(scRNAb1, ndims=20, reduction="pca") 
pc.num=1:20
scRNAb1 <- FindNeighbors(scRNAb1, dims = pc.num) 
scRNAb1 <- FindClusters(scRNAb1, resolution = 1.0)

scRNAb1<-BuildClusterTree(scRNAb1)
install.packages('ape')
PlotClusterTree(scRNAb1)

scRNAb1 = RunTSNE(scRNAb1, dims = pc.num)
embed_tsne <- Embeddings(scRNAb1, 'tsne')
write.csv(embed_tsne,'embed_tsne.csv')
plot1 = DimPlot(scRNAb1, reduction = "tsne") 

plot1

DimPlot(scRNAb1, reduction = "tsne",label = TRUE) 

ggsave("tSNE.pdf", plot = plot1, width = 8, height = 7)

scRNAb1 <- RunUMAP(scRNAb1, dims = pc.num)
embed_umap <- Embeddings(scRNAb1, 'umap')
write.csv(embed_umap,'embed_umap.csv') 
plot2 = DimPlot(scRNAb1, reduction = "umap") 
plot2

二、重点函数解释

1.Read10X

Read10X() 函数从 10X 读取 cellranger流程的输出，返回唯一分子识别（UMI）计数矩阵。该矩阵中的值代表在每个细胞（列）中检测到的每个特征（即基因；行）的分子数。

2.CreateSeuratObject

使用计数矩阵创建一个 Seurat 对象。该对象可作为一个容器，同时包含单细胞数据集的数据（如计数矩阵）和分析（如 PCA 或聚类结果）。

3.PercentageFeatureSet()

PercentageFeatureSet()函数计算线粒体QC指标，该函数计算源自一组特征的计数百分比。
nFeature_RNA: 可以理解为被检测到的基因数
nCount_RNA: 可以理解为被测序到的转录本的数量
R语言正则表达式常用符号：
图片[1]-单细胞分析（一）：矩阵数据的处理-吃了吃了
图片[2]-单细胞分析（一）：矩阵数据的处理-吃了吃了

4.NormalizeData()

从数据集中移除不需要的细胞后，下一步就是对数据进行归一化处理。默认情况下，我们采用全局缩放归一化方法 "LogNormalize"（对数归一化），将每个细胞的特征表达测量值按总表达量归一化，再乘以缩放因子（默认为 10,000），然后对结果进行对数转换。归一化值存储在 [["RNA"]]@data。

5.FindVariableFeatures()

通过直接模拟单细胞数据固有的均值-方差关系改进了之前的版本，并在 FindVariableFeatures() 函数中实现。默认情况下，会返回每个数据集的 2000 个特征。这些特征将用于下游分析，如 PCA。

6.ScaleData()

应用线性变换（"缩放"），这是 PCA 等降维技术之前的标准预处理步骤。ScaleData() 函数
· 移动每个基因的表达量，使细胞间的平均表达量为 0
· 调整每个基因的表达量，使细胞间的方差为 1
· 这一步在下游分析中给予同等权重，因此高表达基因不会占主导地位
结果存储在 [["RNA"]]@scale.data中
在 Seurat v2 中，可以使用 ScaleData() 函数移除单细胞数据集中不需要的变异源。例如，可以 "回归 "与细胞周期阶段或线粒体污染相关的异质性。在 Seurat v3 中，ScaleData()仍然支持这些特征，即


pbmc <- ScaleData(pbmc, vars.to.regress = "percent.mt")
pbmc <- ScaleData(pbmc, vars.to.regress = "percent.mt")
pbmc <- ScaleData(pbmc, vars.to.regress = "percent.mt")

7.RunPCA()

决定数据的维度：
JackStrawPlot() 函数提供了一种可视化工具，用于比较每个 PC 的 p 值分布与均匀分布（虚线）。显著 "的 PC 将显示出低 p 值特征的强烈富集（虚线上方的实心曲线）。在这种情况下，在前 10-12 个 PC 之后，显著性会急剧下降。
另一种启发式方法是生成 "肘图"：根据每个主成分所解释的方差百分比对主成分进行排序（ElbowPlot() 函数）。可以观察到 PC9-10 周围出现了一个 "弯头"，表明前 10 个 PC 中捕捉到了大部分真实信号。

8.FindNeighbors/FindClusters

这些方法将细胞嵌入一个图结构–例如 K 近邻（KNN）图，在具有相似特征表达模式的细胞之间划出边，然后尝试将该图划分为高度相互关联的 "准小区 "或 "社区"。
与 PhenoGraph 一样，首先根据 PCA 空间中的欧氏距离构建一个 KNN 图，然后根据两个单元的局部邻域中的共享重叠（Jaccard 相似性）来细化它们之间的边权重。这一步骤使用 FindNeighbors() 函数执行，并将之前定义的数据集维度（前 10 个 PCs）作为输入。
为了对细胞进行聚类，我们接下来会应用模块化优化技术，如卢万算法（默认值）或 SLM，对细胞进行迭代聚类，目的是优化标准模块化函数。FindClusters() 函数实现了这一过程，并包含一个分辨率参数，用于设置下游聚类的 "粒度"，数值越大，聚类数量越多。我们发现，将该参数设置在 0.4-1.2 之间，通常能为 3K 左右的单细胞数据集带来良好的结果。对于更大的数据集，最佳分辨率通常会提高。可以使用 Idents() 函数找到聚类。

本文遵循知识共享许可 CC-BY-SA 4.0 协议，转载请标明转自“吃了吃了”博客，地址：https://blog.ineuro.net。

THE END