What is meant by this?: "NOTE: though the clusters match the R output, the cluster names are shuffled"

[jolespin]

First off, thank you so much for creating this package. I've been needing something like this and wrote a wrapper with rpy2 but will use this instead from now on.

I am noticing some inconsistencies: (1) The clustering looks the same but the colors from WGCNA seem to repeat themselves. Is this a WGCNA issue or an artifact of your note:

dynamicTreeCut contains methods for detection of clusters in hierarchical clustering dendrograms. "NOTE: though the clusters match the R output, the cluster names are shuffled"

(2) Can you elaborate on what you mean by the quote above? I keep getting clusters that overlap when I plot them ...

Here is the link to the dataframe: https://drive.google.com/file/d/1vp_jx8CfD90bvFcS6sbWN59U-_DQa48L/view?usp=sharing

Here's my Rcode:

library(dynamicTreeCut)
library(fastcluster)
library(WGCNA)

# Read in dataframe
read_dataframe = function(path, sep="\t") {
  df = read.table(path, sep=sep, row.names=1, header = TRUE, check.names=FALSE)
  return(df)
}
df_adj = read_dataframe("~/adj.tsv")

# Convert to dissimilarity
df_dism = 1 - df_adj

# Compute hierarchical clustering linkage
Z = hclust(as.dist(df_dism), method="ward.D2")

# Cut the dendrogram
treecut_output = cutreeDynamic(
  dendro=Z, 
  method="hybrid", 
  distM=df_dism, 
  minClusterSize = 10,
  deepSplit=2,
)

# Plot dendrogram
plotDendroAndColors(
  dendro=Z,
  colors=treecut_output,
)

Here is my python representation using the same parameters:

[Linvill]

First of all, also from my side, thank you very much for putting this package together! I am intending to use the package for the clustering of pico-earthquakes based on waveform similarity.

To get to know the package, I am currently trying to replicate your example here @jolespin. Assuming that your "R-generated solution" above is correct, I am a little confused, because it seems that your first (black) cluster seems to be different, than the one in the "Python-generated solution". Also in my replicated representation using this package, the black cluster does not appear as you have it in your "R-generated solution" (see below).

Trying to figure out what the cause for that could be, I may just think of the linkage method used in R ("ward.D2") compared to scipy ("ward"). Ultimately, I belief its not the cause because absolute linkage values (y-axis in the above plots) do match. Would anybody have an idea on the cause for this mismatch?

On further, more general question: Where are the leafs, which are not assigned to clusters because "minClusterSize = 10" is too large? Are these the black (R-generated solution) or red (Python-generated solution) leafs?

Thanks!