1. 读入单个样本
for (samp in samples){
prefix <- paste0("Data_",samp)
sample.cr.dir <- paste0("Cellranger/",
samp,"/FF/outs/")
seurat_data <- Load10X_Spatial(
data.dir = sample.cr.dir,# 默认filtered_feature_bc_matrix.h5文件为输入
assay = "Spatial",
slice = samp,
filter.matrix = TRUE,
to.upper = FALSE)
#给cellid增加样本名，避免合并时barcode重复，不加的话Seurat会自动给barcode加后缀避免重复
# seurat_data <- RenameCells(seurat_data, add.cell.id = prefix)
seurat_data <- seurat_data[,seurat_data$nCount_Spatial>0]
seurat_data$orig.ident <- samp
seurat_data <- SCTransform(seurat_data,assay = "Spatial",verbose = FALSE) #合并样本时必须都有SCT这个assay才能合并
seurat_data <- FindVariableFeatures(seurat_data) #合并后的数据hvg来自单样本hvg数据
assign(prefix,seurat_data)
}
2. 合并数据
if (length(samples)>1){
other <- ls(pattern="Data_")[-1]
m <- function(x){eval(as.name(x))}
data <- merge(x = get(ls(pattern="MergeData_")[1]),
y = c(lapply(X = other, function(x) m(x))))
}
data$Sample <- data$orig.ident
merged.hvg <- c(lapply(X=ls(pattern="Data_"),function(x) VariableFeatures(x)))
VariableFeatures(data) <- unique(merged.hvg)
DefaultAssay(data) <- "SCT"
这里补充一点更新：合并空间数据后hvg怎么处理？
从github上参考作者回复，有个函数来生成整体的hvg
Yes, we do recommend running SCTransform on each object separately. For selecting variable features, you could either take an intersection or select using SelectIntegrationFeatures and then set VariableFeatures(merged_object) <- my_integration_features.
https://github.com/satijalab/seurat/issues/5761
Select integration features — SelectIntegrationFeatures • Seurat
object.list=list("T0"=t0,"T1"=t1,"T2"=t2,"T3"=t3)
hvgs <- SelectIntegrationFeatures(
object.list,
nfeatures = 2000,
verbose = TRUE,
fvf.nfeatures = 2000
)
VariableFeatures(data) <- unique(hvgs )
Tips
（1）合并后的数据如何删除部分样本
selected_sample_names <- c("Sample1","Sample2","Sample3","Sample4","Sample5","Sample6")
sample6 <- subset(sample7,cells = rownames(sample7@meta.data[sample7$Sample %in% selected_sample_names,]))
sample6@images <- sample6@images[selected_sample_names]
sample6[["SCT"]]@SCTModel.list <- sample6[["SCT"]]@SCTModel.list[1:6]#需要手动先看一下哪个model list是空的
（2）HE图片背景不干净如何修改后替换
需要注意⚠️修图后保存的时候图片的长宽要和原图一致
可使用图片预览工具，上方打开“工具-修改尺寸“
image.png
library(png)
#将修改好的图片读入
new_image <- readPNG(paste0("../HEimage_adjusted/",samp,".tissue_lowres_image.png"))
# 从image替换掉
visium_data[["Sample1"]]@image= new_image
（3）删除部分spot
有时候会有少数spot脱离组织，散落在片子空白区域，我们不想将其纳入分析。
-
可以使用10x软件loupe browser读取cloupe文件，手动选取要删除的spot，将其输出保存未csv文件：
delete_spot.csv - 对样本rds中对应spot进行删除处理
del_spots <- read.csv("delete_spot.csv")
delspots <- del_spots$Barcode
sample1.f <- sample1[,setdiff(colnames(sample1),delspots)]
