Introduction
Overview
Here we provide a worked example of a 'simple' discovery analysis workflow, where the entire process (data prep, clustering, dimensionality reduction, cluster annotation, plotting, summary data, and statistical analysis) is contained within a single script. The 'simple' workflow is most suitable for fast analysis of small datasets. For larger or more complex datasets, or datasets with multiple batches, we recommend the general discovery workflow, where the data preparation, batch alignment, clustering/dimensionality reduction, and quantitative analysis are separated into separate scripts. The demo dataset used for this worked example are cells extracted from mock- or virally-infected mouse brains, measured by flow cytometry.
Strategy
The 'simple' and 'general' discovery workflows are designed to facilitate the analysis of large and complex cytometry datasets using the Spectre R package. We've tested up to 30 million cells in a single analysis session so far. The workflow is designed to get around the cell number limitations of tSNE/UMAP. The analysis starts with clustering with FlowSOM – which is fast and scales well to large datasets. The clustered data is then downsampled, and dimensionality reduction is performed with tSNE/UMAP. This allows for visualisation of the data, and the clusters present in the dataset. Once the possible cell types in the datasets have been explored, the clusters can be labelled with the appropriate cellular identities. Finally, we can use the clusters/populations to generate summary statistics (expression levels, frequencies, total counts etc), which allows us to create graphs and heatmaps, facilitating statistical analysis.
Multiple samples
To analyse multiple samples, all the files must be imported into the one analysis session. This allows cells from each session to be clustered and analysed together, and allows us to examine the differential expression of markers, or the differences in cell proportions, between experimental groups.
Batch alignment
The 'simple' discovery workflow does not include any batch alignment steps. If batch correction needs to be applied, we recommend using the general discovery workflow.
Citation
If you use Spectre in your work, please consider citing Ashhurst TM, Marsh-Wakefield F, Putri GH et al. (2020). bioRxiv. 2020.10.22.349563. To continue providing open-source tools such as Spectre, it helps us if we can demonstrate that our efforts are contributing to analysis efforts in the community. Please also consider citing the authors of the individual packages or tools (e.g. CytoNorm, FlowSOM, tSNE, UMAP, etc) that are critical elements of your analysis work. We have provided some generic text that you can use for your methods section with each protocol and on the 'about' page.
Sample methods blurb
Here is a sample methods blurb for this workflow. You may need to adapt this text to reflect any changes made in your analysis.
Computational analysis of data was performed using the Spectre R package (Ashhurst et al., 2020), with instructions and source code provided at https://github.com/ImmuneDynamics/spectre. Samples were initially prepared in FlowJo, and the population of interest was exported as raw value CSV files. Arcsinh transformation was performed on the data in R using a co-factor of 15 to redistribute the data on a linear scale and compress low end values near zero. The dataset was then merged into a single data.table, with keywords denoting the sample, group, and other factors added to each row (cell). The FlowSOM algorithm (Van Gassen et al., 2015) was then run on the merged dataset to cluster the data, where every cell is assigned to a specific cluster and metacluster. Subsequently, the data was downsampled and analysed by the dimensionality reduction algorithm Uniform Manifold Approximation and Projection (UMAP) (McInnes, Healy, Melville, 2018) for cellular visualisation.
Note
- If you haven't installed Spectre, please visit our Spectre installation page.
- If you aren't familiar with using RStudio or Spectre, please see our RStudio and Spectre basic tutorials.
Software and R script preparation
Software: for instructions downloading R, RStudio, and Spectre, please see this section on the home page.
Analysis script: Please visit https://github.com/ImmuneDynamics/Spectre, and download the repository:
You can then find the 'simple discovery workflow' script here:
Create a folder for your experiment, and place the script in that folder.
Data preparation and export from FlowJo (or similar)
Export the population of interest (POI) from your files.
Please see this page for detailed instructions on exporting data for Spectre. When using fluorescence data, please ensure you are exporting the data from compensated channels, indicated by 'Comp-Channel Name' (e.g. Comp-B515). Feel free to include any other relevant parameters as well (FSC, SSC, time etc). We recommend exporting CSV 'scale' value data, or alternatively exporting FCS files. These represent the untransformed data values. You are also welcome to use the CSV 'channel' value data, which uses a form of 'binning' to transform the data onto a linear distribution – please read this page for a more detailed explanation on CSV channel values.
Data folder
Create a folder within your experiment folder called 'data', and place the exported files there (see this page for more info)..
Metadata folder
Create a folder within your experiment folder called 'metadata', and place a 'sample.details.csv' file there (see this page for more info).
1. Load packages and set directories
Open the analysis script in RStudio and open the simple discovery workflow script.
########################################################################################################## #### 1. Load packages, and set working directory ##########################################################################################################
.
Load the Spectre and other required libraries
Running library(Spectre) will load the Spectre package. We can then use package.check() to see if the standard dependency packages are installed, and package.load() to load those packages.
### Load libraries
library(Spectre)
Spectre::package.check() # Check that all required packages are installed
Spectre::package.load() # Load required packages
.
Set 'PrimaryDirectory'
Initially, we will set the location of the script as 'PrimaryDirectory'. We'll use this as a sort of 'home page' for where our analysis is going to be performed – including where to find our input data, metadata, and where our output data will go.
### Set PrimaryDirectory
dirname(rstudioapi::getActiveDocumentContext()$path) # Finds the directory where this script is located
setwd(dirname(rstudioapi::getActiveDocumentContext()$path)) # Sets the working directory to where the script is located
getwd()
PrimaryDirectory <- getwd()
.
Set 'InputDirectory'
Next we need to set the location of the 'data' folder – where our samples for analysis are stored. In this example they are stored in a sub-folder called 'data'.
Setting directories
If you aren't sure how to navigate directories in R, check out our brief introduction to R tutorial.
### Set 'input' directory
setwd(PrimaryDirectory)
setwd("data/")
InputDirectory <- getwd()
setwd(PrimaryDirectory)
.
Set 'MetaDirectory'
We need to set the location of the 'metadata' folder. This is where we can store a CSV file that contains any relevant metadata that we want to embed in our samples. In this example, it is located in a sub-folder called 'metadata'.
### Set 'metadata' directory
setwd(PrimaryDirectory)
setwd("metadata/")
MetaDirectory <- getwd()
setwd(PrimaryDirectory)
.
Create 'OutputDirectory'
We need to create a folder where our output data can go once our analysis is finished. In this example we will call this 'Output_Spectre'.
### Create output directory
dir.create("Output_Spectre", showWarnings = FALSE)
setwd("Output_Spectre")
OutputDirectory <- getwd()
setwd(PrimaryDirectory)
2. Read and prepare data
########################################################################################################## #### 2. Import and prep data ##########################################################################################################
.
Read in data
To begin, we will change our working directory to 'InputDirectory' and list all the CSV files in that directory – these should be the sample CSV files.
### Import data
setwd(InputDirectory)
list.files(InputDirectory, ".csv")
We can then read in all of our samples (in this example, one CSV file per sample) into a list called 'data.list'. Spectre uses the data.table framework to store data, which reads, writes, and performs operations on data very quickly.
data.list <- Spectre::read.files(file.loc = InputDirectory,
file.type = ".csv",
do.embed.file.names = TRUE)
By default, the read.files() function will generate some other variables, which you can review, by running the do.list.summary() function.
The 'name.table' variable is a table of all the column names for all of your samples (one row per sample, one column per column name). If all of the column names are matching, then this table should be a repeating pattern. If it has been jumbled, then some of your samples have columns that don't appear in other samples. The 'ncol.check' and 'nrow.check' are simple tables indicating the number or columns and rows in each sample.
### Check the data
check <- do.list.summary(data.list)
check$name.table # Review column names and their subsequent values
check$ncol.check # Review number of columns (features, markers) in each sample
check$nrow.check # Review number of rows (cells) in each sample
You can review the first 6 rows of the first sample in your data using the following:
data.list[[1]]
.
Merge data.tables
Once the metadata has been added, we can then merge the data into a single data.table using do.merge.files(). By default, columns with matching names will be aligned in the new table, and any columns that are present in some samples, but not others, will be added and filled with 'NA' for any samples that didn't have that column initially.
### Merge data
cell.dat <- Spectre::do.merge.files(dat = data.list)
Once the data has been merged, we can review the data:
cell.dat
NK11 CD3 CD45 Ly6G CD11b B220 CD8a Ly6C CD4 FileName FileNo
1: 42.3719 40.098700 6885.08 -344.7830 14787.30 -40.2399 83.7175 958.7000 711.0720 CNS_Mock_01 1
2: 42.9586 119.014000 1780.29 -429.6650 5665.73 86.6673 34.7219 448.2590 307.2720 CNS_Mock_01 1
3: 59.2366 206.238000 10248.30 -1603.8400 19894.30 427.8310 285.8800 1008.8300 707.0940 CNS_Mock_01 1
4: 364.9480 -0.233878 3740.04 -815.9800 9509.43 182.4200 333.6050 440.0710 249.7840 CNS_Mock_01 1
5: 440.2470 40.035200 9191.38 40.5055 5745.82 -211.6940 149.2200 87.4815 867.5700 CNS_Mock_01 1
---
169000: 910.8890 72.856100 31466.20 -316.5570 28467.80 -7.7972 -271.8040 12023.7000 1103.0500 CNS_WNV_D7_06 12
169001: -10.2642 64.188700 45188.00 -540.5140 22734.00 202.4110 -936.4920 4188.3300 315.9400 CNS_WNV_D7_06 12
169002: -184.2910 -9.445650 11842.60 -97.9383 17237.00 123.4760 -219.9320 8923.4000 -453.4640 CNS_WNV_D7_06 12
169003: 248.3860 229.986000 32288.20 -681.1630 19255.80 -656.0540 -201.5880 10365.7000 61.6765 CNS_WNV_D7_06 12
169004: 738.9810 95.470300 46185.10 -1004.6000 22957.80 -661.6280 72.3356 9704.4700 -31.8532 CNS_WNV_D7_06 12
.
Read in sample metadata
### Read in metadata
setwd(MetaDirectory)
meta.dat <- fread("sample.details.csv")
meta.dat
3. Arcsinh transformation
Before we perform clustering etc, we need to meaningfully transform the data. For more information on why this is necessary, please see this page.
Data transformations
If you have imported CSV (channel value) files exported from FlowJo, then no data transformations are required, and you can skip all of the arcsinh transformation steps and proceed straight to adding the metadata. More information on the FCS, CSV scale, and CSV channel value file types can be found here.
First, check the column names of the dataset.
##########################################################################################################
#### 3. Data transformation
##########################################################################################################
setwd(OutputDirectory)
dir.create("Output 1 - transformed plots")
setwd("Output 1 - transformed plots")
### Arcsinh transformation
as.matrix(names(cell.dat))
[,1] [1,] "NK11" [2,] "CD3" [3,] "CD45" [4,] "Ly6G" [5,] "CD11b" [6,] "B220" [7,] "CD8a" [8,] "Ly6C" [9,] "CD4" [10,] "FileName" [11,] "FileNo"
The columns we want to apply arcsinh transformation to are the cellular columns – column 1 to column 9. We can specify those columns using the code below.
to.asinh <- names(cell.dat)[c(1:9)]
to.asinh
Define the cofactor we will use for transformation. As a general recommendation, we suggest using cofactor = 15 for CyTOF data, and cofactor between 100 and 1000 for flow data (we suggest 500 as a starting point). Here is a quick comparison figure showing how different co-factors compare to bi-exponential transformations performed on an LSR-II. For more detailed information on this choice, and for approaches where different cofactors for different columns might be required, see this page.
.
In this worked example we will use a cofactor of 500.
cofactor <- 500
You can also choose a column to use for plotting the transformed result – ideally something that is expressed on a variety of cell types in your dataset.
plot.against <- "Ly6C_asinh"
Now we need to apply arcsinh transformation to the data in those columns, using a specific co-factor.
cell.dat <- do.asinh(cell.dat, to.asinh, cofactor = cofactor)
transformed.cols <- paste0(to.asinh, "_asinh")
We can then make some plots to see if the arcsinh transformation is appropriate
for(i in transformed.cols){
make.colour.plot(do.subsample(cell.dat, 20000), i, plot.against)
}
Check the plots and see if you are happy with the transformation. For more detailed guidance, see this page.
If happy, the proceed with analysis. Otherwise, go back to the merging of the data.list (to create cell.dat) and try with another co-factor.
4. Add sample metadata and set preferences
########################################################################################################## #### 4. Add metadata and set some preferences ##########################################################################################################
We also want to read in and attach some sample metadata, to aid with our analysis. We can set our working directory to MetaDirectory and read in the CSV.
setwd(MetaDirectory)
### Metadata
meta.dat <- fread("sample.details.csv")
meta.dat
Once we have the metadata read into R, we will select only the columns we want to add to our dataset. In this example we only want to include use first four columns (Filename, Sample, Group, and Batch). 'Filename' will be used to for matching between cell.dat and meta.dat, and the other three columns will be the information that gets added to cell.dat
sample.info <- meta.dat[,c(1:4)]
sample.info
counts <- meta.dat[,c(2,5)]
counts
Now we can add this information to cell.dat. Essentially, the file names are listed in the metadata table, and we can use that to add any listed metadata in the table to the corresponding files in data.list.
cell.dat <- do.add.cols(cell.dat, "FileName", sample.info, "Filename", rmv.ext = TRUE)
We can review the data to ensure the metadata has been correctly embedded.
cell.dat
Check the column names.
### Define cellular columns
as.matrix(names(cell.dat))
[,1] [1,] "NK11" [2,] "CD3" [3,] "CD45" [4,] "Ly6G" [5,] "CD11b" [6,] "B220" [7,] "CD8a" [8,] "Ly6C" [9,] "CD4" [10,] "FileName" [11,] "FileNo" [12,] "NK11_asinh" [13,] "CD3_asinh" [14,] "CD45_asinh" [15,] "Ly6G_asinh" [16,] "CD11b_asinh" [17,] "B220_asinh" [18,] "CD8a_asinh" [19,] "Ly6C_asinh" [20,] "CD4_asinh" [21,] "Sample" [22,] "Group" [23,] "Batch"
Specify columns that represent cellular features (in this case, the arcsinh transformed data, defined by "markername_asinh").
cellular.cols <- names(cell.dat)[c(12:20)]
as.matrix(cellular.cols)
Additionally, specify the columns that will be used to generate cluster and tSNE/UMAP results. Columns that are not specified here will still be analysed, but won't contributed to the generation of clusters. There are a couple of strategies to take here: use all cellular columns for clustering to looks for all possible cell types/states, or use only stably expressed markers to cluster stable phenotypes, which can then be examined for changes in more dynamic markers. For more guidance, see this page.
cluster.cols <- names(cell.dat)[c(12:20)]
as.matrix(cluster.cols)
Specify sample, group, and batch columns
exp.name <- "CNS experiment"
sample.col <- "Sample"
group.col <- "Group"
batch.col <- "Batch"
Additionally, we want to specify the downsample targets for dimensionality reduction. This influences how many cells will be shown on a tSNE/UMAP plot, and we are specifying the number of cells per group to downsample to. Check for the number of cells (rows) in each group:
### Subsample targets per group
data.frame(table(cell.dat[[group.col]])) # Check number of cells per sample.
Var1 Freq 1 Mock 66992 2 WNV 102012
You can then specify the number to downsample to in each group. These must be lower than the total number of cells in each group, and must be provided in the order the groups appear above). In this example we want 2000 cells from 'mock' and 20,000 cells from 'WNV', to reflect the number of cells present in each group.
sub.targets <- c(2000, 20000) # target subsample numbers from each group
sub.targets
5. Clustering and dimensionality reduction
########################################################################################################## #### 5. Clustering and dimensionality reduction ##########################################################################################################
.
We can run clustering using the run.flowsom function. In this case we can define the number of desired metaclusters manually, with the meta.k argument (in this case we have chosen 8). This can be increased or decreased as required. Typically, overclustering is preferred, as multiple clusters that represent a single cellular population can always be annotated as such. Subsequently, we can write the clustered dataset to disk.
setwd(OutputDirectory)
dir.create("Output - clustering")
setwd("Output - clustering")
### Clustering
cell.dat <- run.flowsom(cell.dat, cluster.cols, meta.k = 8)
fwrite(cell.dat, "clustered.data.csv")
We can then run dimensionality reduction on a subset of the data, allow us to visualise the data and resulting clusters. In this case we have used run.umap, though other options are available, including run.fitsne and run.tsne. As before, this subsampled dataset with DR coordinates is saved to disk.
### Dimensionality reduction
cell.sub <- do.subsample(cell.dat, sub.targets, group.col)
cell.sub <- run.umap(cell.sub, cluster.cols)
fwrite(cell.sub, "clustered.data.DR.csv")
We can visualise the DR data to asses which clusters represent cellular populations
### DR plots
make.colour.plot(cell.sub, "UMAP_X", "UMAP_Y", "FlowSOM_metacluster", col.type = 'factor', add.label = TRUE)
make.multi.plot(cell.sub, "UMAP_X", "UMAP_Y", cellular.cols)
We can also generate some multi plots to compare between experimental groups or batches.
make.multi.plot(cell.sub, "UMAP_X", "UMAP_Y", "FlowSOM_metacluster", group.col, col.type = 'factor')
.We can also produce expression heatmaps to help guide our interpretation of cluster identities.
### Expression heatmap
exp <- do.aggregate(cell.dat, cellular.cols, by = "FlowSOM_metacluster")
make.pheatmap(exp, "FlowSOM_metacluster", cellular.cols)
6. Annotate clusters
########################################################################################################## #### 6. Annotate clusters ##########################################################################################################
Review the cluster labels and marker expression patterns, so you can annotate the clusters. This annotation is optional, as all subsequent steps can be performed on the 'clusters' instead of the 'populations'. Here we can create a list of population names, and then specify which clusters make up that population (e.g. CD4 T cells are contained within cluster '3').
setwd(OutputDirectory)
dir.create("Output 3 - annotation")
setwd("Output 3 - annotation")
### Annotate
annots <- list("CD4 T cells" = c(3),
"CD8 T cells" = c(2),
"NK cells" = c(1),
"Neutrophils" = c(8),
"Infil Macrophages" = c(4),
"Microglia" = c(5,6,7))
Once the annotation list is created, we can switch the list into a table format to annotate our data.
annots <- do.list.switch(annots)
names(annots) <- c("Values", "Population")
setorderv(annots, 'Values')
annots
Values Population 1: 1 NK cells 2: 2 CD8 T cells 3: 3 CD4 T cells 4: 4 Infil Macrophages 5: 5 Microglia 6: 6 Microglia 7: 7 Microglia 8: 8 Neutrophils
Using the do.add.cols function, we can add the population names to the corresponding clusters.
### Add annotations
cell.dat <- do.add.cols(cell.dat, "FlowSOM_metacluster", annots, "Values")
cell.dat
fwrite(cell.dat, "Annotated.data.csv")
cell.sub <- do.add.cols(cell.sub, "FlowSOM_metacluster", annots, "Values")
cell.sub
fwrite(cell.sub, "Annotated.data.DR.csv")
Subsequently, we can visualise the population labels on a UMAP plot.
make.colour.plot(cell.sub, "UMAP_X", "UMAP_Y", "Population", col.type = 'factor', add.label = TRUE)
.
make.multi.plot(cell.sub, "UMAP_X", "UMAP_Y", "Population", group.col, col.type = 'factor')
We can also generate an expression heatmap to summarise the expression levels of each marker on our populations.
### Expression heatmap
rm(exp)
exp <- do.aggregate(cell.dat, cellular.cols, by = "Population")
make.pheatmap(exp, "Population", cellular.cols)
### Write FCS files
setwd(OutputDirectory)
setwd("Output 3 - annotation")
dir.create('FCS files')
setwd('FCS files')
write.files(cell.dat,
file.prefix = exp.name,
divide.by = sample.col,
write.csv = FALSE,
write.fcs = TRUE)
7. Summary data, graphs, statistics
########################################################################################################## #### 7. Summary data ##########################################################################################################
Here we can create 'summary' data for our experiment. This involves calculating the percentage of each population in each sample, along with the corresponding cell counts if the information is available. In addition, we calculate the MFI for selected markers on each population in each sample.
First, set the working directory, and select which columns we will measure the MFI of. In this case, CD11b_asinh and Ly6C_asinh.
setwd(OutputDirectory)
dir.create("Output 4 - summary data")
setwd("Output 4 - summary data")
### Setup
variance.test <- 'kruskal.test'
pairwise.test <- "wilcox.test"
comparisons <- list(c("Mock", "WNV"))
comparisons
grp.order <- c("Mock", "WNV")
grp.order
We can also specify which columns we wish to measure MFI levels on.
### Select columns to measure MFI
as.matrix(cellular.cols)
dyn.cols <- cellular.cols[c(5,8)]
dyn.cols
Use the new create.sumtable function to generate summary data – a data.table of samples (rows) vs measurements (columns).
### Create summary tables
sum.dat <- create.sumtable(dat = cell.dat,
sample.col = sample.col,
pop.col = "Population",
use.cols = dyn.cols,
annot.cols = c(group.col, batch.col),
counts = counts)
Once the summary data has been generated, we can review it and select which columns to plot. In each case, the column names (i.e. name of each summary measure) are structured as 'MEASURE TYPE -- POPULATION'. This provides a useful structure, as we can use regular expression searches to split the name into just the MEASURE TYPE or POPULATION segment.
### Review summary data
sum.dat
as.matrix(names(sum.dat))
[,1] [1,] "Sample" [2,] "Group" [3,] "Batch" [4,] "Percent of sample -- CD4 T cells" [5,] "Percent of sample -- CD8 T cells" [6,] "Percent of sample -- Infil Macrophages" [7,] "Percent of sample -- Microglia" [8,] "Percent of sample -- Neutrophils" [9,] "Percent of sample -- NK cells" [10,] "Cells per sample -- CD4 T cells" [11,] "Cells per sample -- CD8 T cells" [12,] "Cells per sample -- Infil Macrophages" [13,] "Cells per sample -- Microglia" [14,] "Cells per sample -- Neutrophils" [15,] "Cells per sample -- NK cells" [16,] "MFI of CD11b_asinh -- CD4 T cells" [17,] "MFI of CD11b_asinh -- CD8 T cells" [18,] "MFI of CD11b_asinh -- Infil Macrophages" [19,] "MFI of CD11b_asinh -- Microglia" [20,] "MFI of CD11b_asinh -- Neutrophils" [21,] "MFI of CD11b_asinh -- NK cells" [22,] "MFI of Ly6C_asinh -- CD4 T cells" [23,] "MFI of Ly6C_asinh -- CD8 T cells" [24,] "MFI of Ly6C_asinh -- Infil Macrophages" [25,] "MFI of Ly6C_asinh -- Microglia" [26,] "MFI of Ly6C_asinh -- Neutrophils" [27,] "MFI of Ly6C_asinh -- NK cells"
Specify which columns we want to plot.
plot.cols <- names(sum.dat)[c(4:27)]
plot.cols
Reorder the data such that sample appear in the specify group order.
### Reorder summary data
sum.dat <- do.reorder(sum.dat, group.col, grp.order)
sum.dat[,c(1:3)]
.
Violin/scatter plots
We can use the run.autograph function to create violin/scatter plots with embedded statistic – one per population/measurement type.
### Autographs
for(i in plot.cols){
measure <- gsub("\\ --.*", "", i)
measure
pop <- gsub("^[^--]*.-- ", "", i)
pop
make.autograph(sum.dat,
x.axis = group.col,
y.axis = i,
y.axis.label = measure,
grp.order = grp.order,
my_comparisons = comparisons,
Variance_test = variance.test,
Pairwise_test = pairwise.test,
title = pop,
subtitle = measure,
filename = paste0(i, '.pdf'))
}
.
Heatmaps
We can also create a global heatmap show the z-score of each population/measurement type against each sample.
### Create a fold change heatmap
## Z-score calculation
sum.dat.z <- do.zscore(sum.dat, plot.cols)
## Group
t.first <- match(grp.order, sum.dat.z[[group.col]])
t.first <- t.first -1
t.first
## Make heatmap
make.pheatmap(sum.dat.z,
sample.col = sample.col,
plot.cols = paste0(plot.cols, '_zscore'),
is.fold = TRUE,
plot.title = 'Z-score',
dendrograms = 'column',
row.sep = t.first,
cutree_cols = 3)
8. Output session info
########################################################################################################## #### 8. Output session info and FCS files ##########################################################################################################
Create "Output-info" directory and save session data
For the final step of our setup, we want to record the session info our R session, and save this in a folder we'll call "Output-info".
### Session info and metadata
setwd(OutputDirectory)
dir.create("Output - info", showWarnings = FALSE)
setwd("Output - info")
sink(file = "session_info.txt", append=TRUE, split=FALSE, type = c("output", "message"))
session_info()
sink()
write(cellular.cols, "cellular.cols.txt")
write(cluster.cols, "cluster.cols.txt")
