很多情况下，我们需要将多个sample dataset合并在一起，然后进行接下来的分析。Seurat v4官方提供了可以用于整合数据集的函数：FindIntegrationAnchors()和IntegrateData()。
1.导入数据，设置SeuratObject
首先读入数据，然后创建SeuratObject对象。
1
2
3
4
5
6
7
8
9
10
11
12
13
| library(Seurat)
data1 <- Read10X(data.dir = "D:/000 MyWork/000 MyProject/000 scVariants/sample1",
gene.column = 1)
data2 <- Read10X(data.dir = "D:/000 MyWork/000 MyProject/000 scVariants/sample2",
gene.column = 1)
data1 <- CreateSeuratObject(counts = data1,project = "sample1",
min.cells = 3,min.features = 200,
assay = "RNA")
data2 <- CreateSeuratObject(counts = data2,project = "sample2",
min.cells = 3,min.features = 200,
assay = "RNA")
|
接下来分别对两个数据各自进行Normalize（每个都要各自进行normalize），并鉴定variable features,最后选择两个datasets中repeatedly variable features进行integration。
1
2
3
4
5
6
7
| data.list <- c(data1,data2)
data.list <- lapply(X = data.list, FUN = function(x) {
x <- NormalizeData(x)
x <- FindVariableFeatures(x, selection.method = "vst", nfeatures = 2000)
})
features <- SelectIntegrationFeatures(object.list = data.list)
|
2.进行整合
然后，使用FindIntegrationAnchors()函数确定Anchor，这个函数以SeuratObject函数作为输入，然后再利用IntegrateData()函数进行整合。
1
2
3
4
| data.anchors <- FindIntegrationAnchors(object.list = data.list,
anchor.features = features)
data.combined <- IntegrateData(anchorset = data.anchors)
|
这样整合之后的数据只包括了前面鉴定到的features信息，后续的scale等等分析都仅基于这些features，默认是2000个，因此数据会小很多。
3.整合后的分析
接下来就可以进行后续的分析了。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
| DefaultAssay(data.combined) <- "integrated"
data.combined <- ScaleData(data.combined, verbose = FALSE)
data.combined <- RunPCA(data.combined, npcs = 30, verbose = FALSE)
data.combined <- FindNeighbors(data.combined, reduction = "pca", dims = 1:20)
data.combined <- FindClusters(data.combined, resolution = 0.5)
data.combined <- RunUMAP(data.combined, reduction = "pca", dims = 1:20)
p1 <- DimPlot(data.combined, reduction = "umap")
p2 <- DimPlot(data.combined, reduction = "umap", label = TRUE, repel = TRUE)
p1
p2
|
这种整合方法对于较大的数据集，比较消耗内存和时间，了解到还有另外一种整合方法：Harmony，之后再说。
参考：https://satijalab.org/seurat/articles/integration_introduction.html