R语言superheat包：聚类树枝着色/加矩形及树状图复用需求

Great questions—let's break down how to solve both of your superheat needs step by step, leveraging its integration with ggdendro and R's clustering tools:

Superheat uses ggdendro under the hood for dendrograms, so we can manually replicate its clustering logic, customize the dendrogram, and pass it back into the plot.

Step 1: Replicate superheat's Column Clustering #

First, we need to mirror how superheat generates the column dendrogram when col.dendrogram = TRUE and scale = TRUE:

library(superheat)
library(ggdendro)
library(dendextend)
library(tidyverse)

# Replicate superheat's column scaling and clustering
scaled_mtcars <- scale(mtcars)
col_dist <- dist(t(scaled_mtcars))  # Calculate distance between columns (transpose to use row distance)
col_hclust <- hclust(col_dist)

# Define your cluster count k (use methods like gap statistic or domain knowledge to pick k)
k <- 3
col_clusters <- cutree(col_hclust, k = k)

Step 2: Color Dendrogram Branches #

Use dendextend to color branches by cluster, then convert it to a ggplot object that superheat can use:

# Convert hclust to dendrogram and color branches
col_dend <- as.dendrogram(col_hclust) %>%
  color_branches(k = k)

# Generate ggdendro-compatible data for plotting
dend_data <- dendro_data(col_dend, type = "rectangle")

# Build custom colored dendrogram plot
colored_dend_plot <- ggplot() +
  geom_segment(data = segment(dend_data), aes(x = x, y = y, xend = xend, yend = yend, color = color)) +
  scale_color_identity() +  # Keep the branch colors we defined
  theme_dendro()  # Use ggdendro's minimal theme

# Pass the custom dendrogram to superheat
superheat(mtcars,
          scale = TRUE,
          left.label = "none",
          col.dendrogram = colored_dend_plot,  # Use our colored dendrogram
          legend = FALSE
)

Step 3: Add Cluster Rectangles (Alternative to Branch Coloring) #

If you prefer highlighting clusters with rectangles instead, calculate the x-axis bounds for each cluster and add them to the dendrogram plot:

# Get leaf positions from the dendrogram data
leaf_positions <- dend_data$labels$x
names(leaf_positions) <- dend_data$labels$label

# Calculate x ranges for each cluster
cluster_bounds <- map_df(unique(col_clusters), function(clust) {
  cluster_leaves <- names(col_clusters[col_clusters == clust])
  cluster_x <- leaf_positions[cluster_leaves]
  tibble(
    xmin = min(cluster_x) - 0.5,
    xmax = max(cluster_x) + 0.5,
    ymin = 0,
    ymax = max(dend_data$segment$y),  # Match the dendrogram's maximum height
    cluster = clust
  )
})

# Build dendrogram with cluster rectangles
rect_dend_plot <- ggplot() +
  geom_segment(data = segment(dend_data), aes(x = x, y = y, xend = xend, yend = yend)) +
  geom_rect(data = cluster_bounds, aes(xmin = xmin, xmax = xmax, ymin = ymin, ymax = ymax),
            fill = scales::hue_pal()(k), alpha = 0.2) +  # Semi-transparent rectangles
  theme_dendro()

# Use in superheat
superheat(mtcars,
          scale = TRUE,
          left.label = "none",
          col.dendrogram = rect_dend_plot,
          legend = FALSE
)

To apply the same column order and dendrogram to a new dataset with identical variables, we just need to extract the cluster order from the original hclust object and enforce it on the new data.

Step 1: Extract the Original Column Order #

Pull the ordered column names from the original clustering:

# Get the column order from the original hclust object
original_col_order <- col_hclust$order
original_col_names <- colnames(mtcars)[original_col_order]

Step 2: Apply the Order to the New Dataset #

Use the extracted order to align the new dataset's columns, then pass the original dendrogram to superheat:

# Example new dataset (same variables as mtcars)
set.seed(123)
new_mtcars <- mtcars %>% mutate(across(everything(), ~ .x + rnorm(nrow(mtcars), 0, 0.1)))

# Option 1: Pre-scale and reorder the new data, then plot
scaled_new_mtcars <- scale(new_mtcars)[, original_col_names]
superheat(scaled_new_mtcars,
          left.label = "none",
          col.dendrogram = colored_dend_plot,  # Reuse the colored dendrogram
          legend = FALSE,
          scale = FALSE  # Already scaled manually
)

# Option 2: Let superheat handle scaling, but enforce column order
superheat(new_mtcars,
          scale = TRUE,
          left.label = "none",
          col.dendrogram = colored_dend_plot,
          legend = FALSE,
          column.order = original_col_order  # Force columns to match original cluster order
)

This ensures the new dataset's columns are arranged exactly like the original, and the dendrogram will align perfectly with the heatmap.

内容的提问来源于stack exchange，提问作者Gustavo