In this module, we use the model on single-cell images to clearly demonstrate its application.
Dr. Thomas Walter of the MitoCheck consortium kindly provided the single-cell sample image data. This dataset contains sample single-cell images in the following format:
mitocheck_single_cell_sample_images
│
└───phenotypic_class
│ │
│ └───sample_image_path.png
Because we already extracted the features for these cells in mitocheck_data
, we do not re-extract features from these images in this module.
Instead, features are associated with a single-cell image based on the cell's location metadata (plate, well, frame, x, y).
In correct_15_images.ipynb, we show 15 sample single-cell images that the final model from 2.train_model correctly classifies. Three single-cell images from each of the 5 top performing classes (as determined by F1 score from compiled_F1_scores.tsv) are displayed and their paths are saved in top_5_performing_classes.tsv.
Use the commands below to run the Jupyter notebooks and extract the sample image data:
# Make sure you are located in 6.single_cell_images
cd 6.single_cell_images
# Activate phenotypic_profiling conda environment
conda activate phenotypic_profiling
# Interpret model
bash single_cell_images.sh