R语言 | 使用 ComplexHeatmap 绘制热图，分区并给对角线分区加黑边框

目的：画热图，分区，给对角线分区添加黑色边框
建议直接看0和4。

0. 准备数据

# 安装并加载必要的包
#install.packages("ComplexHeatmap")  # 如果尚未安装
library(ComplexHeatmap)

# 使用 iris 数据集 #data(iris)

# 选择数值列（去掉物种列）
data0 <- iris
rownames(data0)=paste0(iris$Species, 1:nrow(data0))

# data0 <- mtcars 分类效果不好


# 随机抽取30个
set.seed(42)
dat=data0[sample(nrow(data0), 30), 1:4]
#dat=data0

# 计算余弦距离
#install.packages("proxy")      # 如果尚未安装
library("proxy")
distance_matrix <- as.matrix(dist(dat, method = "cosine"))
# 如果不想安装新包，也可以使用默认的欧氏距离：
#distance_matrix <- as.matrix(dist(iris_data, method = "euclidean"))

# 使用相似性绘图 simi=1-dist
similarity=1-distance_matrix

dim(similarity)
[1] 30 30

1. Heatmap 全部分块加黑框

library(circlize)
col_fun = colorRamp2(c(-2, 0, 2), c("green", "white", "red"))
col_fun(seq(-3, 3))

Heatmap(similarity, name = "mat", #col = col_fun,
        row_km = 3, column_km = 3,
        )
# 每个分块绘制黑边框
# When the heatmap is split, layer_fun is applied in every slice.
Heatmap(similarity, name = "mat", #col = col_fun,
        row_km = 3, column_km = 3,
        layer_fun = function(j, i, x, y, width, height, fill) {
            # 全部分块都加黑框
            v = pindex(similarity, i, j)
            #grid.text(sprintf("%.1f", v), x, y, gp = gpar(fontsize = 10))
            str(v)
            grid.rect(gp = gpar(lwd = 2, fill = "transparent"))
            if(sum(v > 0)/length(v) > 0.75) {
            }
        })

2. 为对角线分块添加黑边框

Heatmap(similarity, name = "mat",
        #col = c("white", "yellow", "red3"),
        #col = col_fun,
        col =  colorRamp2(c(0.5, 0.75, 1), 
                          c("white", "yellow", "red3")),
        row_km = 3, column_km = 3,
        layer_fun = function(j, i, x, y, width, height, fill, slice_r, slice_c) {
          v = pindex(similarity, i, j)
          #grid.text(sprintf("%.1f", v), x, y, gp = gpar(fontsize = 10))
          if(slice_r == slice_c) {
            grid.rect(gp = gpar(lwd = 4, fill = "transparent", col="black"))
          }
        })

3. 添加列注释

还有一个与 pheatmap 包同名的函数：

annotation_col = data.frame(
  type = data0$Species,
  row.names = rownames(data0)
)[rownames(dat), ,drop=F]
# set colors
ann_colors = list(
  #type = c('setosa'="#ed553b", 'versicolor'="#99b433", 'virginica'="orange")
  type = c('setosa'="violetred1", 
           'versicolor'="turquoise2", 
           'virginica'="blueviolet")
)
# "#ed553b", "#99b433"
#violetred1,turquoise2,

pheatmap(similarity,
         name = "Cosine\nsimilarity",
         main="xx", 
         border_color = NA,
         clustering_method = "ward.D2",
         annotation_col = annotation_col, #set anno for column
         annotation_colors = ann_colors, #set colors
         
         #col = c("white", "yellow", "red3"),
         #col = col_fun,
         col =  colorRamp2(c(0.8, 0.9, 1), 
                  c("white", "yellow", "red3")),
         row_km = 3, column_km = 3,
         layer_fun = function(j, i, x, y, width, height, fill, slice_r, slice_c) {
           v = pindex(similarity, i, j)
           #grid.text(sprintf("%.1f", v), x, y, gp = gpar(fontsize = 10))
           if(slice_r == slice_c) {
             grid.rect(gp = gpar(lwd = 4, fill = "transparent", col="black"))
           }
         })

Bug:

有一个问题：不同次执行，图竟然是不同的，不仅仅是分类的排列顺序问题，而是分类本身也不同了。搜了一下， 竟然受到随机数种子的影响？！固定的数据，固定的参数，每次聚类为什么还要受到随机数影响？不理解！难道非监督的聚类还要人工判断对不对？

比如，对以上最后一个聚类函数，设置不同的随机数种子，结果分别是：

# set.seed(45) #这个随机数竟然影响分类位置？！比如修改随机数种子，图分别为
pheatmap(similarity,
         name = "Cosine\nsimilarity",
         main="xx", 
         border_color = NA,
         clustering_method = "ward.D2",
         annotation_col = annotation_col, #set anno for column
         annotation_colors = ann_colors, #set colors
         
         #col = c("white", "yellow", "red3"),
         #col = col_fun,
         col =  colorRamp2(c(0.8, 0.9, 1), 
                  c("white", "yellow", "red3")),
         row_km = 3, column_km = 3,
         layer_fun = function(j, i, x, y, width, height, fill, slice_r, slice_c) {
           v = pindex(similarity, i, j)
           #grid.text(sprintf("%.1f", v), x, y, gp = gpar(fontsize = 10))
           if(slice_r == slice_c) {
             grid.rect(gp = gpar(lwd = 4, fill = "transparent", col="black"))
           }
         })

原因：使用kmeans聚类，确实是随机数确定初始中心的。不使用kmeans聚类，就不会受到随机数的影响。

4. 层次聚类，对结果分群

• 原来：row_km = 3, column_km = 3, #kmeans确实是种子确定初始中心，结果会随随机数而变化
• 现在：cutree_row=3, cutree_cols=3, #层次聚类是稳定的

pheatmap(similarity,
         name = "Cosine\nsimilarity",
         main="Hierarchical cluster", 
         border_color = NA,
         clustering_method = "ward.D2",
         annotation_col = annotation_col, #set anno for column
         annotation_colors = ann_colors, #set colors
         
         #col = c("white", "yellow", "red3"),
         #col = col_fun,
         col =  colorRamp2(c(0.8, 0.9, 1), 
                           c("white", "yellow", "red3")),
         #row_km = 3, column_km = 3, #kmeans确实是种子确定初始中心
         cutree_row=3, cutree_cols=3, #层次聚类是稳定的
         layer_fun = function(j, i, x, y, width, height, fill, slice_r, slice_c) {
           v = pindex(similarity, i, j)
           #grid.text(sprintf("%.1f", v), x, y, gp = gpar(fontsize = 10))
           if(slice_r == slice_c) {
             grid.rect(gp = gpar(lwd = 4, fill = "transparent", col="black"))
           }
         })

Ref