Question

是否有一种简单的方法可以在根处合并多个hclust对象（或树形图）？

我已尽可能完整地举例说明我的问题。

让我们说我想按区域聚集USArrests，然后联合所有最好的对象，将它们一起绘制在热图中。

USArrests
Northeast <- c("Connecticut", "Maine", "Massachusetts", "New Hampshire", "Rhode Island", 
"Vermont", "New Jersey", "New York", "Pennsylvania")
Midwest <-  c("Illinois", "Indiana", "Michigan", "Ohio",  "Wisconsin", 
    "Iowa", "Kansas", "Minnesota", "Missouri", "Nebraska", "North Dakota", 
    "South Dakota")
South <- c("Delaware", "Florida", "Georgia", "Maryland", "North Carolina", 
           "South Carolina", "Virginia", "West Virginia", 
           "Alabama", "Kentucky", "Mississippi", "Tennessee", "Arkansas", 
           "Louisiana", "Oklahoma", "Texas")
West <- c("Arizona", "Colorado", "Idaho", "Montana", "Nevada", "New Mexico", 
          "Utah", "Wyoming", "Alaska", "California", "Hawaii", "Oregon", "Washington")

h1 <- hclust(dist(USArrests[Northeast,]))
h2 <- hclust(dist(USArrests[Midwest,]))
h3 <- hclust(dist(USArrests[South,]))
h4 <- hclust(dist(USArrests[West,]))

现在我有4个hclust对象（h1到h4）。我通常将它们合并如下：

hc <- as.hclust(merge(merge(merge(
    as.dendrogram(h1), as.dendrogram(h2)), as.dendrogram(h3)), 
    as.dendrogram(h4)))

然后，为了绘制它们，我必须根据hclust对象重新排序矩阵，然后绘图（我添加了一些注释以使绘图更清晰）：

usarr <- USArrests[c(Northeast, Midwest, South, West),]

region_annotation <- data.frame(Region = c(rep("Northeast", length(Northeast)), 
                                rep("Midwest", length(Midwest)),
                                rep("South", length(South)),
                                rep("West", length(West))),
                                row.names = c(Northeast, Midwest, South, West))

pheatmap(usarr, cluster_rows = hc, 
         annotation_row = region_annotation)

总结：除了合并所有单独的hclusts之外，还有一种更简单的方法吗？

Answer 1

要创建合并的hclust对象，您可以在使用<<-创建的自定义环境中安全地使用new.env。

在不使用<<-的情况下，可能还有其他两种方法可以一次创建两个合并对象。希望有人可以对它进行说明。

我尝试使用do.call('merge', list( dendrograms of h1, h2, h3, h4 )。但它没有用，因为hclust需要在顶部有两个分支而不是4个分支。

<强>代码：

library('pheatmap')
myenv <- new.env()
myenv$hc <- as.dendrogram( hclust( dist(USArrests[Northeast,])))
invisible( lapply( list( Midwest, South, West), function(x){
  myenv$hc <<- merge( myenv$hc, as.dendrogram( hclust( dist( USArrests[ x, ]) )) )
  NULL
} ) )
myenv$hc <- as.hclust(myenv$hc)

<强>图形：

pheatmap(usarr, cluster_rows = myenv$hc, 
         annotation_row = region_annotation)

Answer 2

我最终制作了几个函数来更自动地完成这项工作。（在我的版本中，我还添加了对相关“距离”的支持，所以它有点大）

hclust_semisupervised <- function(data, groups, dist_method = "euclidean",
                                  dist_p = 2, hclust_method = "complete") {
    hclist <- lapply(groups, function (group) {
        hclust(dist(data[group,], method = dist_method, p = dist_p), method = hclust_method)
    })
    hc <- .merge_hclust(hclist)
    data_reordered <- data[unlist(groups),]

    return(list(data = data_reordered, hclust = hc))
}

.merge_hclust <- function(hclist) {
    #-- Merge
    d <- as.dendrogram(hclist[[1]])
    for (i in 2:length(hclist)) {
        d <- merge(d, as.dendrogram(hclist[[i]]))
    }
    as.hclust(d)
}

拥有USArrests和区域向量，我这样称呼hclust_semisupervised：

semi_hc <- hclust_semisupervised(USArrests, list(Northeast, Midwest, South, West)

现在绘制热图：

pheatmap(semi_hc$data, cluster_rows = semi_hc$hclust, 
         annotation_row = region_annotation)

合并多个hclust对象（或树形图）

2 个答案: