theislab / deepflow

This code contains the neural network implementation from the nature communication manuscript NCOMMS-16-25447A.

Reconstructing cell cycle and disease progression using deep learning

This code contains the neural network implementation from the nature communication publication https://doi.org/10.1038/s41467-017-00623-3 . A video showing the cell cycle reconstruction is available at: https://www.youtube.com/watch?v=eyWcHIiCazE

Access the data

The data will be shortly hosted at https://data.broadinstitute.org/bbbc/. In the mean time you can download the data here: https://www.dropbox.com/s/tzbhp1skpjsmfsn/CellCycle.zip?dl=0

Running the code.

To reproduce the results from the publication, change the PATH2MXNET variable in generate_record_files.sh to your mxnet home folder and run:

sh generate_record_files.sh

Run the neural network training & prediction:

python3.4 run.py

System Requirements

The results were generated with python3.4 on an Ubuntu 14.04 machine. Additional dependencies:

• mxnet 0.10.0
• numpy 1.12.0
• cv2 3.2.0