Single-Cell Morphological Profiling Reveals Insights into Cell Death

This repository contains analysis scripts for the paper "Single-cell morphological profiling reveals insights into cell death" (https://www.biorxiv.org/content/10.1101/2025.01.15.633042v1).

The repository is structured to facilitate reproducibility and includes scripts for supervised and unsupervised analysis, calculation of key metrics, and generation of publication figures.

---

Repository Structure

• supervised/
  Contains scripts for supervised classification tasks using single-cell and aggregated profiles.

  • Supervised_analysis.ipynb: Jupyter notebook for evaluation of supervised models.
  • autogluon_classifier_celldeath.py: Script for model training using AutoGluon.

• unsupervised/
  Contains scripts for clustering and dimensionality reduction tasks.

  • Unsupervised_analysis.ipynb: Jupyter notebook for unsupervised analysis, including UMAP and PCA.

• metrics/
  Scripts for calculating perturbation metrics like grit score and e-distance.

  • grit_script.py: Script to calculate grit scores.
  • run_etest.py: Script to calculate e-distances and perform permutation tests.
  • etest_grit_analysis.ipynb: Notebook for analyzing and visualizing metric results.

• visualization/
  Scripts for generating figures and visualizations.

  • visualize_attention_Celldeath.py: Script to generate attention maps and related plots.

• config/
  Configuration files for reproducibility.

  • config_finetune_unmasked.yaml: Configuration for training DINO models. Note: We do not provide all imaging data here.

• data_prep/
  Data preparation files for different feature extractors.

  • data_prep.py: File with main normalization and plate concatenation functions
  • CellProfiler_FeatureExtraction.py: Function for feature selection and QC of CellProfiler features.
  • CellProfiler_QC.py: Quality control functions, needs to be linked to database (proprietary internal structure).
  • utils.py: Utils for data_prep.py

---

DINO Training

For DINO training, we adapted the codebase from DINO4Cells. Please refer to their repository for detailed instructions. The checkpoints to our model can found in FigShare.

---

CellViewer

To obtain representative cell images shown in Fig. 3, Suppl. Fig. S5–S7, S9, S10, and S14, we used the CellViewer tool. The code for CellViewer is available at CellViewer.

---

Data Availability

Extracted features can be found on FigShare. Provided are normalised profiles. Aggregation and analysis steps can be found in the scripts. Grit scores and e-distance result can be found on FigShare.

Citation

If you use this code or data, please cite:

@article{frey_single-cell_2025,
  title   = {Single-cell morphological profiling reveals insights into cell death},
  author  = {Frey, Benjamin and Holmberg, David and Byström, Petter and Bergman, Ebba and Georgiev, Polina and Johansson, Martin and Hennig, Patrick and Rietdijk, Jonne and Rosén, Dan and Carreras-Puigvert, Jordi and Spjuth, Ola},
  journal = {bioRxiv},
  year    = {2025},
  doi     = {10.1101/2025.01.15.633042},
  url     = {https://www.biorxiv.org/content/early/2025/01/19/2025.01.15.633042}
}