Seurat包基本分析实战

作者: 小贝学生信 | 来源:发表于2020-10-02 00:56 被阅读0次

    一、背景知识

    二、大致思路

    (1)根据GEO号下载表达矩阵,以及meta信息;
    (2)根据meta信息筛选tumor cell 的表达矩阵;
    (3)根据表达矩阵构建Seurat对象,以及添加分组信息、质控;
    (4)归一化→降维→聚类;
    (5)组图

    三、具体实现

    1、下载数据

    https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE84465

    1-1
    1-2
    1-3

    2、导入数据

    a <- read.table("GSE84465_GBM_All_data.csv.gz")
    a[1:4,1:4]
    dim(a) 
    #[1] 23465  3589
    #原始矩阵有3598个cell,23465个gene
    b <- read.table("SraRunTable.txt",sep = ",",header = T)
    dim(b)
    b[1:4,1:4]
    #[1] 3589   33
    
    • 通过观察,表达矩阵a:3589个cell(name感觉是不同信息的组合),23465个gene(symbol ID)
    • 通过观察meta b了解到cell name是主要根据每个细胞的plate_id与well组合而成;并且有tissue、patient_id两种分类方式在后面分析会用到。

    3、筛选目标类型cell

    new.b <- data.frame(names=paste0("X",b$plate_id,".",b$Well),
    #names即为a的colname
                        tissue=b$TISSUE, #tumor二分类
                        group=b$Patient_ID) #样本分类
    
    new.b <- new.b[match(colnames(a),new.b$names),]
    #使顺序一致,以便后期的筛选
    table(new.b$names==colnames(a))
    rownames(new.b) <- colnames(a)
    
    index <- new.b$tissue=="Tumor"
    a_filt <- a[,index]
    dim(a_filt)
    #挑选到2343个tumor cell 表达矩阵
    

    b$Patient_ID有四类,说明这些细胞是从四个病人的病灶组织中取的。

    4、构建Seurat对象,以及质控、可视化

    group=data.frame(group=new.b[index,3],
                     row.names = colnames(a_filt))
    
    library("Seurat")
    scRNA = CreateSeuratObject(counts=a_filt,
                               meta.data = group,
                               min.cells = 3, min.features = 50)
    scRNA[["percent.mt"]] <- PercentageFeatureSet(scRNA, pattern = "^MT-")
    #标记线粒体基因
    head(scRNA@meta.data)
    summary(scRNA@meta.data)
    #结果显示没有线粒体基因,因此感觉过滤也没有意义
    pctMT=5
    scRNA <- subset(scRNA, subset = percent.mt < pctMT)
    
    • 通过结果可以看到仅过滤掉一个细胞;
    • 但有趣的是我是按照文献的三个标准过滤的,而文献是过滤掉了194个cell。
    • 从这里开始,分析结果就与文献产生了差异。但目前也不知道是为什么。
    4-1

    可视化三个图

    #图A:观察不同组cell的counts、feature分布
    col.num <- length(levels(scRNA@meta.data$group))
    p1_1 <- VlnPlot(scRNA,
            features = c("nFeature_RNA","nCount_RNA"),
            group.by = "group",
            cols =rainbow(col.num)) +
      theme(legend.position = "none") +
      labs(tag = "A")
    #图B:nCount_RNA与对应的nFeature_RNA关系
    library(ggplot2)
    p1_2 <- FeatureScatter(scRNA, feature1 = "nCount_RNA", feature2 = "nFeature_RNA",
                   group.by = "group",pt.size = 1.3) +
      labs(tag = "B")
    

    5、挑选hvg高变基因

    #标准化
    scRNA <- NormalizeData(scRNA, normalization.method = "LogNormalize", scale.factor = 10000)
    #归一化
    scRNA <- ScaleData(scRNA, features = (rownames(scRNA)))
    scRNA <- FindVariableFeatures(scRNA, selection.method = "vst", nfeatures = 1500) 
    top10 <- head(VariableFeatures(scRNA), 10) 
    plot1 <- VariableFeaturePlot(scRNA) 
    p1_3 <- LabelPoints(plot = plot1, points = top10, repel = TRUE, size=2.5) +
      theme(legend.position = c(0.1,0.8)) +
      labs(tag = "C")
    #标记top10 hvg
    
    library(patchwork)
    p1_1 | p1_2 | p1_3
    
    上三个子图

    6、降维

    #默认使用前2000个hvg降维处理
    scRNA <- RunPCA(scRNA, features = VariableFeatures(scRNA)) 
    p2_1 <- DimPlot(scRNA, reduction = "pca", group.by="group")+
      labs(tag = "D")
    
    #挑选主成分
    scRNA <- JackStraw(scRNA,reduction = "pca")
    scRNA <- ScoreJackStraw(scRNA,dims = 1:20)
    p2_2 <- JackStrawPlot(scRNA,dims = 1:20, reduction = "pca") +
      theme(legend.position="bottom") +
      labs(tag = "E")
    p2_3 <- ElbowPlot(scRNA, ndims=20, reduction="pca") 
    #结果显示可挑选前20个pc
    
    p2_1| (p2_2 | p2_3)
    
    中三个子图

    7、聚类

    pc.num=1:20
    scRNA <- FindNeighbors(scRNA, dims = pc.num) 
    # dims参数,需要指定哪些pc轴用于分析;这里利用上面的分析,选择20
    scRNA <- FindClusters(scRNA, resolution = 0.4)
    table(scRNA@meta.data$seurat_clusters)
    #  0   1   2   3   4   5   6   7   8   9  10  11  12 
    #393 380 281 271 177 140 122 108  72  51  41  35  18 
    scRNA = RunTSNE(scRNA, dims = pc.num)
    p3_1 <- DimPlot(scRNA, reduction = "tsne") +
      labs(tag = "E")
    
    • 如上结果,可知分为13个clust;
    • 寻找每个clust的marker gene,这里使用的是FindAllMarkers()函数
    diff.wilcox = FindAllMarkers(scRNA)
    #大概2-3min
    head(diff.wilcox)
    library(tidyverse)
    all.markers = diff.wilcox %>% select(gene, everything()) %>%
      subset(p_val<0.05 & abs(diff.wilcox$avg_logFC) > 0.5)
    top20 = all.markers %>% group_by(cluster) %>% top_n(n = 20, wt = avg_logFC)
    top20
    #0-9共10个cluster,每个cluster选取top10高变基因
    top20 = CaseMatch(search = as.vector(top20$gene), match = rownames(scRNA)) 
    length(top20)
    length(unique(sort(top20)))
    #这里选取的是wilcox方法挑选的差异基因
    p3_2 <- DoHeatmap(scRNA, features = top20, group.by = "seurat_clusters")
    

    此处的结果也是与原文差别比较大的地方。均是对每个clust寻找top20 marker gene。但是原文使用的limma包识别,去重后仅有96个gene,而我自己尝试的或还有227个,相差比较大。
    The differential analysis identified 8,025 marker genes. The top 20 marker genes of each cell cluster are displayed in the heatmap. A total of 96 genes are listed beside of the heatmap after omitting the same top marker genes among clusters. The colors from purple to yellow indicate the gene expression levels from low to high.

    p3_1 | p3_2
    
    下二个子图

    8、总图合并

    p <- (p1_1 | (p1_2 | p1_3) ) /
      ((p2_1| p2_2 | p2_3) /
         (p3_1 | p3_2))
    ggsave("my_try.pdf", plot = p, width = 15, height = 18) 
    
    my_try.pdf

    相关文章

      网友评论

        本文标题:Seurat包基本分析实战

        本文链接:https://www.haomeiwen.com/subject/bcbvuktx.html