cytoVAE - Extended Variational Auto-Encoder for Single-Cell Representation Learning

This repository contains the code of a tensorflow implementation of the auto-encoder model presented in the following publication:

[1] Capturing Single-Cell Phenotypic Variation via Unsupervised Representation Learning ; Maxime W. Lafarge, Juan C. Caicedo, Anne E. Carpenter, Josien P.W. Pluim, Shantanu Singh, Mitko Veta ; MIDL 2019; PMLR 102:315-325

Content

• training_cytoVAE.py: Main script to run the training procedure.
• models/VAEPlus/: Directory containing the model components (architecture, loss definition, optimizers, training procedure).
• dataManagers/: Abstract module to handle a dataset and generate training batches of images.
• exp_cytoVAE_demo/: Abstract module to define an experimental setup (imported by the main script).

Requirements

The current code was developed and tested with the following configuration:

• python 3.4.5
• tensorflow-gpu 1.14.0
• numpy 1.16.4

Notes

• Dataset and code to import images need to be added in order for the script to run (changes to dataManagers and exp_cytoVAE_demo must also be adapted to the user's project).