Feature extraction with DeepProfiler
This folder contains the config files to run DeepProfiler feature extraction using a pretrained model.
Follow the Installation guide.
Copy the tiff images to the inputs/images/ folder.
Copy the csv files containing the cell locations to the inputs/locations
folder.
Download the pretrained model and move it to outputs/efn_pretrained/checkpoint/efficientnetb0_notop.h5
Extract features using the profile option:
$ python3 -m deepprofiler --exp efn_pretrained --root ./ --config jump.json profileRun the aggregation script to build well-based profiles:
$ python3 utils/build_profiles.pyIt will generate the efn_pretrained.parquet file with the following
structure:
| Metadata_Plate | Metadata_Well | 0 | 1 | 2 | ... | 6397 | 6398 | 6399 | |
|---|---|---|---|---|---|---|---|---|---|
| 0 | BR00116996 | A01 | -0.145242 | -0.0976836 | -0.110939 | ... | -0.188828 | -0.0994691 | 1.11783 |
| 1 | BR00116996 | A02 | -0.16149 | -0.0870829 | -0.130767 | ... | -0.185635 | -0.11242 | 1.04221 |
| 2 | BR00116996 | A03 | -0.146293 | -0.0802944 | -0.123345 | ... | -0.180474 | -0.105045 | 1.01936 |
| 3 | BR00116996 | A04 | -0.153452 | -0.0818477 | -0.133035 | ... | -0.187012 | -0.103294 | 0.992679 |
| 4 | BR00116996 | A05 | -0.167078 | -0.0862786 | -0.134154 | ... | -0.187428 | -0.109548 | 1.11078 |