CellCognition Explorer - Deep Learning Extension

CellCognition Explorer Deep Learning learning Extension enables the unsupervised learning of cellular features directly based on image data. Only the rough bounding-box is required.

The Extension can be operated by a graphical user interface or by command line.

Prerequisites

To run cedl.py, please unzip the demo data

• cellH5-file : CecogExCedlDemo_plate_1.ch5
• full-mapping : CecogExCedlDemo_plate_1_mapping_full.txt
• encode-mapping : CecogExCedlDemo_plate_1_mapping_encode.txt (contains only a subset of positions) into the "data" folder of the cedl command-line tool.

The graphical user interface

Using installation based on docker

Downlaod ...

The command line

Help pages

$ python cedl.py --help

$ python cedl.py train --help

$ python cedl.py encode --help

Training

$ python cedl.py --im_size 60 --verbose train --learner nesterov --autoencoder_architecture c16.5r_p2_c32.3r_p2_d256.1r_d64.0s ../data/CecogExCedlDemo_plate_1.ch5 ../data/CecogExCedlDemo_plate_1_mapping_full.txt

This will create an deep learning autoencoder model called "CecogExCedlDemo_plate_1_c16.5r_p2_c32.3r_p2_d256.1r_d64.0s". This name is needed in the enocding step:

Encoding

$ python cedl.py --im_size 60 --verbose encode CecogExCedlDemo_plate_1_c16.5r_p2_c32.3r_p2_d256.1r_d64.0s ../data/CecogExCedlDemo_plate_1.ch5 ../data/CecogExCedlDemo_plate_1_mapping_encode.txt

Will create "CecogExCedlDemo_plate_1_c16.5r_p2_c32.3r_p2_d256.1r_d64.0s.hdf"

Visualization and Novelty detection

Open CellCognition Explorer GUI and load "CecogExCedlDemo_plate_1_c16.5r_p2_c32.3r_p2_d256.1r_d64.0s.hdf"