cytofkit: run with an example

Load the Package

require("cytofkit")

Open the package help page to take a first look of the pakcage:

?cytofkit

Options for using the package

cytofkit package provides three ways to employ the workforce of this package:

GUI with function cytof_tsne_densvm_GUI
The core function cytof_tsne_densvm
Step-by-step

The most simple way to use the cytofkit package is to either use the GUI or the core function. Both provide almost all the options for this CyToF analysis pipeline. Start the GUI with a simple command cytof_tsne_densvm_GUI(), you will find all the options there on the control pannel, click the ? buttion to get the explanation of each parameter. Launching one analysis simply like this: Choose the input fcs files from the directory where you store the data; select the markers from the auto-generated marker list; choose the directory to save the output; give a base name as a prefix of the names of result files; select a merging method if you have multiple files then you can submit you analysis. Depends on the size of your data, it will take some time to run the analysis. Once it's done, all the results will be generated under you result directory. The core function has these parameters in a command way, you can define these inputs as in the GUI in the arguments of the function cytof_tsne_densvm, check the help page ?cytof_tsne_densvm to check the arguments you need to change.

The GUI and the core function is all you need if you want to make use of the cytofkit package easily. For deep customizing the analysis, the cytofkit package also provides a step-by-step guide for the analysis pipeline.

Step-by-step guide

Due to the usually big size of the fcs data, we just use one small size file here as a demo, you can easily expand the analysis to multiple fcs files.

Load the data

This step is done using function fcs_lgcl_merge. The expression data of one or multiple fcs files is extracted, and be merged into one data matrix. Then a transformation method(usually is logicle transformation) is applied to the combined expression matrix. Row names with sample name and sample id are added to the matrix to label the cells.

dir <- system.file('extdata',package='cytofkit')
files <- list.files(dir, pattern='.fcs$', full=TRUE)
paraFile <- list.files(dir, pattern='.txt$', full=TRUE)
parameters <- as.character(read.table(paraFile, sep = "\t", header = TRUE)[, 1])
#exprs <- fcs_lgcl_merge(fcsFile = files, markers = parameters, lgclMethod = "fixed", mergeMethod = "all")

Dimension reduction

Dimension reduction is implemented in function cytof_dimReduction, methods like isomap, pca and tsne are provided as options. tsne is the recommended and default method.

?cytof_dimReduction   ## check the help page
#transformed <- cytof_dimReduction(exprs, method = "tsne")

Clustering

A Density-based clustering aided by support Vector Machine (function densVM_cluster) is provided to automate subset detection from the transformend map. Of course other clustering methods can be applied instead.

?densVM_cluster   ## check the help page
#clusterRes <- densVM_cluster(transformed, exprs)

Cluster annotation using scatter plot and heat map

Take a look at the cluster reuslts:

clusters <- clusterRes[[2]]
head(clusters)

##             tsne_1      tsne_2 cluster
## b1c3a_1  7.6808488   8.2444303       4
## b1c3a_2 -2.0409126   9.7981320       4
## b1c3a_3 22.5045863 -16.7973727       9
## b1c3a_4 -4.0605419 -27.1562123       3
## b1c3a_5 18.3696005   0.6531961       8
## b1c3a_6 -0.9903563  10.0439368       4

Scatter plot

Scatter plot visualize the cell points with colour indicating their assigned clusters and point shape representing their belonging samples. Use cluster_plot to visulalize the cluster results:

cluster_plot(clusters, title = "Demo cluster", point_size = 2)

plot of chunk unnamed-chunk-8

If the input contains multiple fcs files, can use the cluster_gridPlot to visualize the cluster of different samples.

Heatmap plot

Heat map visualizing the mean expression of every marker in every cluster is generated with no scaling on the row or column direction with function clust_mean_heatmap. Hierarchical clustering is added using Euclidean distance and complete agglomeration method.

exprs_cluster <- data.frame(exprs, cluster = clusters[, 3])
clust_statData <- clust_state(exprs_cluster, stat = "mean")
clust_mean_heatmap(clust_statData[[1]], baseName = "Demo")

## NULL

plot of chunk unnamed-chunk-9

If the input contains multiple fcs files, can use the clust_percentage_heatmap to visualize the percentage of cells of each cluster in each sample.