Single-cell imaging-AI based chromatin biomarkers for proton therapy efficacy in tumor patients using liquid biopsies
The repository contains the code used to run the analyses presented in our publication:
Chromatin organization of PBMCs reflects the presence of tumor signals
The code has been developed and executed on a HP Z4 workstation running Ubuntu 20.04.5 LTS with a Intel(R) Xeon(R) W-2255 CPU with 3.70 GHz, 128GB RAM and Python v3.8 installed.. Note that the code can also be run for machines with less available RAM.
To install the code, please clone the repository and install the required software libraries and packages listed in the requirements.txt file:
git clone https://github.com/GVS-Lab/immune_cell_project.git
conda create --name icp python=3.8
conda activate icp
pip install -r requirements.txt
The data to rerun the results can be obtained from the authors upon request and will be made publicly available upon publication.
The full preprocessing pipeline including the image segmentation in 2D and 3D as well as the feature extraction and (if applicable) the cell type classification can be run via the following commands:
python run.py --config config/full_pipeline/config_pbmc_full.yml
python run.py --config config/full_pipeline/config_pbmc_full_marker.yml
Note that the config file needs to be adjusted such that the correct file locations are specified. While all other parameters are recommended to be kept the channel parameters should also be adjusted to reflect the channels present in the images that are to be preprocessed.
The above steps can be skipped, if the preprocessed data is used which can be obtained from the authors (see Data resources)
All results presented in the main and supplemental figures can be obtained by running the respective jupyter notebooks located in jupyter/notebooks/figures. To run those notebooks, please start the jupyter server via
jupyter notebook
navigate to the notebook and run all cells.
If you use any of the code or resources provided here please make sure to reference the required software libraries if needed and also cite our work: