如何从heatmaply层级聚类第1阶段簇提取测试样本t1的相似样本

Hey there! Let's walk through exactly how to extract those similar samples to t1 and their features using hclust, dist, and a deep dive into the merge property you're confused about. I’ll keep this practical with code examples so you can follow along.

Step 1: Align Your Clustering with heatmaply #

First, critical note: Make sure your manual clustering matches what heatmaply used. Heatmaply defaults to:

• Distance metric: euclidean (for rows/samples)
• Clustering method: complete linkage

If you changed these in heatmaply, mirror those parameters in your dist() and hclust() calls. Let’s start with a reproducible example using your dataset structure:

# Load your dataset (replace with your actual S)
# Example: S has rows as samples, columns as features, with t1 included
set.seed(123)
S <- data.frame(
  Feature1 = rnorm(10),
  Feature2 = rnorm(10),
  Feature3 = rnorm(10)
)
rownames(S) <- paste0("Sample", 1:10)
S <- rbind(S, t1 = c(0.5, 0.6, 0.4)) # Add test sample t1

# Replicate heatmaply's clustering setup
dist_matrix <- dist(S, method = "euclidean") # Row-wise distance (samples)
hc <- hclust(dist_matrix, method = "complete") # Match heatmaply's linkage

Step 2: Understand hclust's merge Property #

The merge matrix is the key here—it records every step of the clustering process:

• Each row represents one merge event (from first to last)
• Negative numbers refer to original sample indices (e.g., -3 means the 3rd sample in your dataset)
• Positive numbers refer to previously merged clusters (e.g., 5 means the cluster formed in the 5th merge step)

Since you saw t1 form a cluster in the first stage, we need to find the earliest merge that includes t1.

Step 3: Locate t1's First Merge #

First, get t1's index in your dataset:

t1_idx <- which(rownames(S) == "t1")

Now scan the merge matrix to find the first row that includes -t1_idx (the negative index represents t1 as an individual sample):

# Find the first merge involving t1
target_merge <- NULL
for (i in 1:nrow(hc$merge)) {
  if (-t1_idx %in% hc$merge[i, ]) {
    target_merge <- hc$merge[i, ]
    break
  }
}

Step 4: Extract All Samples in the Initial Cluster #

If your initial cluster is just t1 + one other sample, target_merge will have two negative numbers (e.g., -11 and -7). If it’s t1 + a small pre-formed cluster (positive number), we need a quick recursive function to unpack all samples in that sub-cluster:

# Recursive function to unpack all samples from a merge entry
get_all_cluster_samples <- function(merge_entry, full_merge_matrix) {
  samples <- c()
  for (entry in merge_entry) {
    if (entry < 0) {
      samples <- c(samples, abs(entry)) # Add individual sample index
    } else {
      # Recursively unpack the pre-formed cluster
      samples <- c(samples, get_all_cluster_samples(full_merge_matrix[entry, ], full_merge_matrix))
    }
  }
  return(unique(samples))
}

# Get all indices of samples in t1's initial cluster
cluster_sample_indices <- get_all_cluster_samples(target_merge, hc$merge)

Step 5: Grab the Sample Names & Features #

Now it’s straightforward to extract the data you need:

# Get similar sample names
similar_sample_names <- rownames(S)[cluster_sample_indices]

# Extract their feature values
similar_sample_features <- S[cluster_sample_indices, ]

# Output the results
cat("Samples similar to t1 (initial cluster):", paste(similar_sample_names, collapse = ", "), "\n")
print(similar_sample_features)

Quick Note on order Property #

You mentioned confusion about hc$order—this is just the order of samples as they appear in the heatmap’s dendrogram. It’s used for plotting, but we don’t need it here to extract the initial cluster.

内容的提问来源于stack exchange，提问作者Aslan