ℹ️ Beta Version Available: A beta version with compatibility for Google Colab and newer versions of torch and scvi-tools is available on the beta-colab branch. Install it with pip install celldisect==0.2.0b1.

🧬 Overview

CellDISECT (Cell DISentangled Experts for Covariate counTerfactuals) is a powerful causal generative model that enhances single-cell analysis by:

• 🔍 Disentangling Variations: Separates covariate variations at test time
• 🧪 Counterfactual Predictions: Learns to make accurate counterfactual predictions
• 🎯 Flexible Fairness: Achieves flexible fairness through expert models for each latent space
• 🔬 Enhanced Discovery: Captures both covariate-specific information and novel biological insights

📚 Documentation

Visit our comprehensive documentation for:

• Detailed API reference
• Step-by-step tutorials
• Best practices and examples
• Advanced usage guides

🚀 Quick Start

Prerequisites

We recommend using Anaconda/Miniconda. Create and activate a new environment:

conda create -n CellDISECT python=3.9
conda activate CellDISECT

Installation

1. Install PyTorch (tested with pytorch 2.1.2 and cuda 12):

conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia

2. Install CellDISECT:

# Via pip (stable version)
pip install celldisect

# Or via GitHub (latest development version)
pip install git+https://github.com/Lotfollahi-lab/CellDISECT

Optional Dependencies

pip install \
    --extra-index-url=https://pypi.nvidia.com \
    cudf-cu12==24.4.* dask-cudf-cu12==24.4.* cuml-cu12==24.4.* \
    cugraph-cu12==24.4.* cuspatial-cu12==24.4.* cuproj-cu12==24.4.* \
    cuxfilter-cu12==24.4.* cucim-cu12==24.4.* pylibraft-cu12==24.4.* \
    raft-dask-cu12==24.4.* cuvs-cu12==24.4.*

pip install rapids-singlecell

pip install -U "jax[cuda12_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

📖 Tutorials & Examples

Basic Tutorials

Tutorial	Description	Links
Basic Training	Learn how to train CellDISECT and make counterfactual predictions using the Kang dataset

Perturbation Prediction

Tutorial	Description	Links
Perturbation Prediction	Predict gene expression under seen, unseen, and combinatorial perturbations using predefined embeddings (GenePT, ESM)

Advanced Applications

Tutorial	Description	Links
Latent Space Analysis	Explore combinations of CellDISECT latent spaces for erythroid subset inference
Double Counterfactual	Advanced tutorial recreating Scenario 2 counterfactual on the Eraslan dataset

🧪 Perturbation Prediction

CellDISECT supports perturbation prediction using predefined gene embeddings (e.g. GenePT, ESM). This enables prediction for unseen perturbations and combinatorial perturbations.

import numpy as np
from celldisect import CellDISECT, perturbation_metrics

# Store predefined embeddings in adata.uns
adata.uns['pert_embeddings'] = gene_embeddings  # dict: gene_name -> np.ndarray

# Setup with perturbation support
CellDISECT.setup_anndata(
    adata,
    layer='counts',
    categorical_covariate_keys=['cell_type', 'perturbation'],
    perturbation_key='perturbation',
    perturbation_embedding_key='pert_embeddings',
)

# Train the model
model = CellDISECT(adata, n_latent_shared=32, n_latent_attribute=32)
model.train(max_epochs=200)

# Predict seen, unseen, or combinatorial perturbations
x_ctrl, x_true, x_pred = model.predict_perturbation(
    adata,
    perturbation='GeneA+GeneB',
    source_perturbation='ctrl',
    cats=['cell_type', 'perturbation'],
    perturbation_key='perturbation',
)

# Evaluate
metrics = perturbation_metrics(x_pred.numpy(), x_true.numpy(), x_ctrl.numpy())

See the perturbation prediction tutorial for a full walkthrough.

🤝 Contributing

We welcome contributions! Please see our contributing guidelines for details on how to:

• Report issues
• Submit bug fixes
• Propose new features
• Submit pull requests

📜 License

This project is licensed under the BSD 3-Clause License - see the LICENSE file for details.

📫 Contact

For questions and support:

• Open an issue
• Visit our documentation

📝 Citation

If you use CellDISECT in your research, please cite our paper:

Megas, S., Amani, A., Rose, A., Dufva, O., Shamsaie, K., Asadollahzadeh, H., Polanski, K., Haniffa, M., Teichmann, S. A., & Lotfollahi, M. (2025). Integrating multi-covariate disentanglement with counterfactual analysis on synthetic data enables cell type discovery and counterfactual predictions. bioRxiv. https://doi.org/10.1101/2025.06.03.657578

@article{Megas2025CellDISECT,
    title={Integrating multi-covariate disentanglement with counterfactual analysis on synthetic data enables cell type discovery and counterfactual predictions},
    author={Megas, Stathis and Amani, Arian and Rose, Antony and Dufva, Olli and Shamsaie, Kian and Asadollahzadeh, Hesam and Polanski, Krzysztof and Haniffa, Muzlifah and Teichmann, Sarah Amalia and Lotfollahi, Mohammad},
    journal={bioRxiv},
    year={2025},
    doi={10.1101/2025.06.03.657578},
    elocation-id={2025.06.03.657578},
    publisher={Cold Spring Harbor Laboratory},
    URL={https://www.biorxiv.org/content/10.1101/2025.06.03.657578v1}
}