LEAP: Layered Ensemble of Autoencoders and Predictors

Predicting gene essentiality and drug response from perturbation screens in preclinical cancer models

Overview

LEAP is a novel ensemble framework designed to improve robustness and generalization in predicting perturbation responses from molecular profiles. By leveraging multiple DAMAE (Data Augmented Masked Autoencoder) representations and LASSO regressors, LEAP consistently outperforms state-of-the-art approaches in predicting gene essentiality and drug responses across diverse biological contexts.

Installation

make install

This will install uv (if needed) and all project dependencies.

Contributing

Run quality checks before committing:

make checks   # Run pre-commit hooks (linting, formatting, type checking)
make tests  # Run tests with coverage

Usage

Clean up temporary files

make clean

Removes all temporary files, caches, and build artifacts.

Clean up data and results

make clean-data

Removes all processed data, trained models, and results. Warning: This action cannot be undone!

Run the complete LEAP pipeline

sh run_pipeline.sh

Runs the full LEAP pipeline end-to-end:

1. Pretrain representations
2. Train regression heads for multiple tasks, models, and seeds
3. Ensemble predictions

The pipeline is configured to run on task 1 with the mae_ps_enet model across 5 different seeds.

Citation

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

@misc{bodinier2025predictinggeneessentialitydrug,
      title={Predicting gene essentiality and drug response from perturbation screens in preclinical cancer models with LEAP: Layered Ensemble of Autoencoders and Predictors},
      author={Barbara Bodinier and Gaetan Dissez and Lucile Ter-Minassian and Linus Bleistein and Roberta Codato and John Klein and Eric Durand and Antonin Dauvin},
      year={2025},
      eprint={2502.15646},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2502.15646},
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Authors

• Barbara Bodinier
• Gaetan Dissez
• Lucile Ter-Minassian
• Linus Bleistein
• Roberta Codato
• John Klein
• Eric Durand
• Antonin Dauvin