Graph Variational Causal Inference

This repository is the official implementation of Predicting Cellular Responses with Variational Causal Inference and Refined Relational Information. The L008 CROP-seq dataset released with our paper can be found here.

@inproceedings{wupredicting,
  title={Predicting Cellular Responses with Variational Causal Inference and Refined Relational Information},
  author={Wu, Yulun and Barton, Rob and Wang, Zichen and Ioannidis, Vassilis N and De Donno, Carlo and Price, Layne C and Voloch, Luis F and Karypis, George},
  booktitle={The Eleventh International Conference on Learning Representations},
  year={2023}
}

Installation

1. Create Conda Environment

conda config --append channels conda-forge
conda create -n gvci-env --file requirements.txt
conda activate gvci-env

2. Install Learning Libraries

• Pytorch 1.13.1
• Pytorch Geometric 2.3.1

  * make sure to install the right versions for your toolkit

3. Install Submodule

git submodule update --init --recursive
pip install -e variational-causal-inference

Preparation

Visit our resource site, download the contents of main/datasets into datasets and the contents of main/graphs into graphs. To process your own dataset, see data-prep branch; to generate your own graph, see graph-prep branch. If using gene relational graphs is not desired, see the repository for Variational Causal Inference. On top of variational causal inference, graphVCI uses the following model architecture to leverage gene relational information.

Run

Once the environment is set up and the contents are prepared, the function call to train & evaluate graphVCI is:

./main.sh &

A list of flags may be found in main.sh and main.py for experimentation with different hyperparameters. The run log and models are saved under *artifact_path*/saves, and the tensorboard log is saved under *artifact_path*/runs.

License

Contributions are welcome! All content here is licensed under the MIT license.