Classification algorithms evaluated against published clustering methods for automated flow cytometry data analysis

Running analysis

• Evaluation data used in the study is unavailable due to patient confidentiality.
• To use your own input data, a pickled pandas Dataframe is required in the format where each row is an FC event, each column is a marker and additional columns label and sample_name indicate the event label and the name of the sample the event is from, respectively.

benchmarking

• benchmarking.py begins the pipeline to run clustering methods on the input data and match outputs to labelled populations using the Hungarian assignment algorithm, then evaluate performance.
• FLOCK requires installation from here.
• flowgrid requires installation from here.
• helper_match_evaluate_multiple.R was made available by Weber and Robinson, 2016.

classification

• classification.py will run Decision Tree, Random Forest and XGBoost classifiers to train and evaluate on input data.

Results

figures

• figures.py allows recreation of the figures in the publication, except for; Fig 6, Fig 7 and Table 2 which would require the publication of confidential patient data.
• figs is the folder where the figures will be saved to.

data

• Data from the study which is required by figure.py to recreate publication figures.
• confusion_matrix contains the confusion matrices generated from each trained classification model.
• f1.pickle contains the results of the evaluation of the classification methods.
• output contains the results and confusion matrices generated from each benchmarked clustering method.