NOTE: This work is outdated, please refer to https://github.com/andkopf/MoESimVAE for the latest results.

CellGAN is a Generative Adversarial Network that aims to learn the cellular heterogeneity from flow cytometry data in an interpretable manner and use it for defining subpopulations.

• Mixture of Gaussian - For proof-of-concept testing
• Bodenmiller Mass Cytometry - More information can be found here. (TODO: Add other datasets?)

NOTE: All commands must be run in a terminal from project root.

The following steps are in reference to running jobs on the Leonhard cluster:

• To train a CellGAN model on the Mixture of Gaussian data, run make gaussian
• To train a CellGAN model on the Bodenmiller data run make bodenmiller

Baseline methods are run for the Bodenmiller Mass Cytometry data. The baseline methods we compare to include (All commands must be run in a terminal from project root):

• FlowSOM: Train FlowSOM by running make flowsom (TODO)
• GMM: Train Gaussian Mixture Model by running make gmm
• XShift: Train XShift by running make xshift (TODO)
• PhenoGraph: Train PhenoGraph by running make phenograph

GitHub: https://github.com/SofieVG/FlowSOM

GitHub: https://github.com/nolanlab/vortex/

GitHub: https://github.com/jacoblevine/PhenoGraph