从dendextend::circlize_dendrogram()的输出中提取簇颜色

Question

我正在尝试提取circlize_dendrogram集群中使用的颜色。下面是一个示例代码：

library(magrittr)
library(dendextend)

cols <- c("#009000", "#FF033E", "#CB410B", "#3B444B", "#007FFF")
dend <- iris[1:40,-5] %>% dist %>% hclust %>% as.dendrogram    

dend <- color_branches(dend, k = 5, col = cols)
dend %<>% set("labels_col", value = cols, k= 5)
dend %<>% set("labels_cex", .8)
dend %<>% set("branches_lwd", 2)

circlize_dendrogram(dend)

​

​

这样就可以使用cutree(dend, k = 5)提取表聚类。是否有一种方法可以根据给定的cols提取树状图中簇的颜色？我需要它来使用grid包在绘图中插入一个图例。

例如:集群1- #009000；集群2- #FF033E；集群3- #CB410B；集群4- #3B444B；集群5- #007FFF.circlize_dendrogram的问题是用于集群的颜色的排序是不同的。

虽然我可以手动完成这个任务，但是如果我能够自动完成的话，它将是有效的。如果我能提取星系团的颜色，那是可能的。

Answer 1

好的，这是一个非常麻烦的解决方案。我相信还有更好的，但这是第一次，所以请原谅我。

其思想是搜索dend对象(在内部是一个列表)，查找相应的元素名称(在本例中仅为数字)，并提取相应的颜色，将其保存在数据帧中，并将其用于图例。

# First we'll extract the elements and corresponding categories...
categories <- cutree(dend, k = 5)

# ... and save them in a data frame
categories_df <- data.frame(elements = as.numeric(names(categories)),
       categories = categories, 
       color = NA)

# now here's a little function that extracts the color for each element
# from the 'dend' object. It uses the list.search() function from the
# 'rlist' package

library(rlist)

extract_color <- function(element_no, dend_obj) {
  dend.search <- list.search(dend_obj, all(. == element_no))
  color <- attr(dend.search[[1]], "edgePar")$col
  return(color)
}

# I use 'dplyr' to manipulate the data
library(dplyr)

categories_df <- categories_df %>% 
  group_by(elements) %>% 
  mutate(color = extract_color(elements, dend))

现在，这给我们提供了以下数据框架：

> categories_df
Source: local data frame [40 x 3]
Groups: elements [40]

   elements categories   color
      (dbl)      (int)   (chr)
1         1          1 #CB410B
2         2          1 #CB410B
3         3          1 #CB410B
4         4          1 #CB410B
5         5          1 #CB410B
6         6          2 #009000
7         7          1 #CB410B
8         8          1 #CB410B
9         9          3 #007FFF
10       10          1 #CB410B
..      ...        ...     ...

我们可以把它总结成一个数据框架，只有类别的颜色。

legend_data <- categories_df %>% 
  group_by(categories) %>% 
  summarise(color = unique(color))

> legend_data
Source: local data frame [5 x 2]

  categories   color
       (int)   (chr)
1          1 #CB410B
2          2 #009000
3          3 #007FFF
4          4 #FF033E
5          5 #3B444B

现在很容易生成图例：

circlize_dendrogram(dend)
legend(-1.05, 1.05, legend = legend_data$categories, fill = legend_data$color, cex = 0.7)

这给了你：

​

​

可以使用cutree(dend, k = 5)确认类别颜色的数字与每个元素的类别相对应。

Answer 2

除了Felix的解决方案外，我还想发表我自己的答案：

library(magrittr)
library(grid)
library(gridExtra)
library(dendextend)

cols <- c("#009000", "#FF033E", "#CB410B", "#3B444B", "#007FFF")
dend <- iris[1:40,-5] %>% dist %>% hclust %>% as.dendrogram    

dend <- color_branches(dend, k = 5, col = cols)
dend %<>% set("labels_col", value = cols, k= 5)
dend %<>% set("labels_cex", .8)
dend %<>% set("branches_lwd", 2)

clust <- cutree(dend, k = 5)
colors <- labels_colors(dend)[clust %>% sort %>% names]
clust_labs <- colors %>% unique

circlize_dendrogram(dend)
grid.circle(x = .95, y = .9, r = .02, gp = gpar(fill = clust_labs[1])) 
grid.circle(x = .95, y = .85, r = .02, gp = gpar(fill = clust_labs[2]))
grid.circle(x = .95, y = .8, r = .02, gp = gpar(fill = clust_labs[3]))
grid.circle(x = .95, y = .75, r = .02, gp = gpar(fill = clust_labs[4]))
grid.circle(x = .95, y = .7, r = .02, gp = gpar(fill = clust_labs[5]))

grid.text(x = .95, y = .9, label = expression(bold(1)), gp = gpar(fontsize = 9, col = "white"))
grid.text(x = .95, y = .85, label = expression(bold(2)), gp = gpar(fontsize = 9, col = "white"))
grid.text(x = .95, y = .8, label = expression(bold(3)), gp = gpar(fontsize = 9, col = "white"))
grid.text(x = .95, y = .75, label = expression(bold(4)), gp = gpar(fontsize = 9, col = "white"))
grid.text(x = .95, y = .7, label = expression(bold(5)), gp = gpar(fontsize = 9, col = "white"))
grid.text(x = .91, y = .8, label = "CLUSTERS", rot = 90, gp = gpar(fontsize = 9))

​

​