Automated clustering methods

In the manuscript Figure 4 details a number of methods that offer potential alternatives to thresholding methods. The manuscript focuses on a method for automating the designation of thresholds. Clustering methods are an alternative and viable way to identify rare subsets and a great deal of effort is being made to overcome some of the difficulties that bar these methods from becoming more widespread.

Because clustering methods are increasingly important in terms of flow cytometry analysis pipelines we document here the exact procedure used to create Figure 4. See the manuscript for further discussion.

Figure 4

Methods

1. Prerequisites
   • Install Python
   • Install matplotlib
   • Install NumPy
   • Install scikit-learn
   • Install py-fcm

1. Create a directory for the scripts and data and move into it e.g.

   ~$ mkdir clustering-thresholds
   ~$ cd clustering-thresholds
   

2. Download and unzip the eqapol-11c-1 fcs files

3. Download the scripts and libraries

   • RunClustering.py
   • MakeClusteringFigure.py
   • EmpiricalCdf.py
   • PositivityLib.py

4. Run the scripts

   ~$ python RunClustering.py
   ~$ python MakeClusteringFigure.py
   

A directory called clustering will be created and all clustering labels as well as figures will be saved within. Once labels are produced for a given clustering method the default is to load them from file. To rerun a clustering method with say a different set of parameters simply remove the appropriate *.npy file from the clustering directory and rerun the scripts.