Seurat Object

基本構造

colnames(so)  # cells
rownames(so)  # genes

ASSAY

DefaultAssay(so) <- "RNA"

Ident

# IDENT 1
Idents(so)
levels(so)
so <- RenameIdents(so, "0"="beta_cell")

# IDENT 2
ids <- c(....)
names(ids) <- levels(so)
so <- RenameIdents(so, ids)

Metadata

so[[]]                                   # 簡略記法
so[["col"]]
so[[]][N]
colnames(so@meta.data)["hoge"] <- "foo"  # 列名変更hoge -> foo: so[[]]ではだめ
so@meta.data$col <- Idents(so)           # 列内容変更

Rename rowname

so@assays$RNA@counts@Dimnames[[1]]      <- c()
so@assays$RNA@data@Dimnames[[1]]        <- c()
so@assays$RNA@scale.data@Dimnames[[1]]  <- c()
so@assays$RNA@meta.features             <- as.data.frame(c())

13 slot

• so@assays
• so@meta.data
• so@active.assay
• so@graphs
• so@reductions
• so@commands

関数出力

CreateSeuratObject

• @meta.data$orig.ident
• @meta.data$nCount_RNA
• @meta.data$nFeature_RNA

NormalizeData

• @assays$RNA@data

FindVariableFeatures

• @assays$RNA@var.features
• @assays$RNA@meta.features

ScaleData

• @assays$RNA@scale.data

RunPCA

• @reductions$pca

FindNeighbors

• @graphs$RNA_nn
• @graphs$RNA_snn

FindClusters

• @meta.data$RNA_snn_res.N
• @meta.data$seurat_clusters

RunUMAP

• @reductions$umap

Subset/Downsampling/Integrate/Merge

SUBSET

sub <- subset(so, idents=c("0", "1"))           # Idents
sub <- subset(so, subset=seurat_clusters!="0")  # meta.data
sub <- subset(so, subset=Ins1>5)                # Gene Exp

"%nin%" <- Negate("%in%")
sub_1 <- subset(sub, subset = seurat_clusters %in% c("4") & Ins2 < 4.5)
sub_2 <- subset(sub, subset = seurat_clusters %nin% c("4"))
sub <- merge(sub_1,sub_2)

DOWNSAMPLING

sub <- so[, sample(colnames(so), size = 3000, replace=F)]

INTEGRATE

so_1 <- NormalizeData(so_1)
so_2 <- NormalizeData(so_2)
so_1 <- FindVariableFeatures(so_1) 
so_2 <- FindVariableFeatures(so_2) 

so <- FindIntegrationAnchors(c(so_1, so_2), dims=1:30)
so <- IntegrateData(so, dims=1:30)

so <- ScaleData(so) 

MERGE

so <- merge(so1, so2)

Visualize

options(repr.plot.width=10, repr.plot.height=10)

DimPlot(so, 
  reduction = "umap", 
  group.by = "meta1",
	label = TRUE,
  pt.size = .8, 
  label.size = 8
)

DimPlot() + NoLegend()
DimPlot() + NoAxes()

DimPlot(sub, 
  cells.highlight = list(
    high = WhichCells(so, expression = Gcg > 6 ),
    low = WhichCells(so, expression = Gcg < 6 & Gcg > 3)
  ),
  cols.highlight = c("blue", "red")
)

VlnPlot(so, features=c("gene1", "gene2"))
VlnPlot(sub, features = c("gene1","gene2"), stack = T, flip = T)

FeaturePlot(so, features=c("gene1", "gene2"))
FeaturePlot(so, features="nCount_RNA")

DotPlot(sub, features = c("Ins2", "Gcg", "Sst"), dot.scale = 10)

DoHeatmap(so, features = c("Ins2", "Gcg", "Sst")) + scale_fill_gradientn()
DoHeatmap(so, features = , size = 0) +  
  theme (axis.text.y  = element_text(size = 18, family = "Helvetica")
  ) + NoLegend()

DEGs

FindMarkers(so, ident.1="0", ident.2=c("1", "2"))

TOP10 HEATMAP

mks <- FindAllMarkers(so, verbose=FALSE)
mks_pos <- mks[mks$avg_log2FC > 0,]
mks_neg <- mks[mks$avg_log2FC < 0,]

top10 <- do.call(rbind, lapply(split(mks_pos, mks_pos$cluster), function(x) {
x[order(x$avg_log2FC, decreasing = TRUE), ][1:10,]}))
top10 <- na.omit(top10)
DoHeatmap(so, top10)

Convert object

Seurat to AnnData

# step1 on R
library(Seurat)
library(SeuratDisk)

so <- readRDS("so.rds")
SaveH5Seurat(so, filename = "so.h5Seurat")
Convert("so.h5Seurat", dest = "h5ad")

# step2 on python
import scanpy as sc
adata = sc.read_h5ad("so.h5ad")

# 名前変更など
adata.obs.seurat_clusters = adata.obs.seurat_clusters.astype("category")
adata.obs.drop(columns="orig.ident", inplace=True)
adata.obs.rename(columns={"seurat_clusters":"clusters"}, inplace = True)

sc.pl.umap(adata, color=['clusters'])

Seurat to Loom

library(Seurat)
library(SeuratDisk)

so <- readRDS("so.rds")
SaveLoom(so, "so.loom")

Alevin Fry to Seurat

library(SingleCellExperiment)
library(fishpond)
library(Seurat)

sce <- loadFry("ALEVIN_FRY_OUTPUT", 
               outputFormat = list("counts" = c("S", "A"),"unspliced" = c("U")))
so <- CreateSeuratObject(counts(sce))
saveRDS(so, "XXX.rds")

Others

I/O

so <- CreateSeuratObject(Read10X(CELLRANGER_DIR), min.cells=3)
so <- readRDS("obj.rds")
saveRDS(so, file="obj.rds")
ggsave("hoge.png", obj)
write.table(obj, "filename", sep=" ")

SUBCLUSTER

so <- RunPCA(so)
so <- FindNeighbors(so, graph.name="grp")
so <- FindClusters(so, graph.name="grp")
so <- RunUMAP(so)

so <- FindSubCluster(so, "1", "grp", subcluster.name="sub", resolution=.5)
  
	# input:  seurat_cluster 1
  # output: so@meta.data$sub 

Rename after doubletfinder

rename <- function(X){
DF = colnames(X[[]])[grepl("DF.classification", colnames(X[[]]))]
n = which(colnames(X[[]]) == DF)
colnames(X@meta.data)[n] = "doublet"
return(X)
}

so <- rename(so)

#@ Make Doublet Table
mktable <- function(X){
x = X[[]]$doublet
y = X[[]]$seurat_clusters

nDouble = tapply(x == "Doublet", y, sum)
nTotal = tapply(x, y, length)
pDouble = sprintf("%.1f", nDouble/nTotal*100)
df = data.frame(doublet_count=nDouble,total_count=nTotal, doublet_percent=pDouble)
return(df)
}

so <- doubletFinder_v3(XXX)
df <- mktable(so)
barplot(as.double(df$doublet_percent), names.arg=rownames(df), las=2)

so_sub <- subset(so, subset = doublet == "singlet")

# IDENT/BAR PLOT
cnt <- table(so@active.ident, so@meta.data$origin)
cnt <- as.data.frame(cnt)
head(cnt)
cnt$cluster <- factor(cnt$cluster, levels = cnt$cluster)
ggplot(cnt, aes(x = Var2, Freq, fill=Var1)) + 
  geom_bar(
    position = "fill", 
    stat     = "identity")