PaSCient: Learning multi-cellular representations of single-cell transcriptomics data enables characterization of patient-level disease states

Setup environment

conda create -n pascient python=3.11

Install

pip install -r requirements.txt

Train a PaSCient model

cd /pascient/scripts/

HYDRA_FULL_ERROR=1 python train.py data.persistent_workers=True data.multiprocessing_context=null data.num_workers=12 experiment=pascient_multilabel paths.tiledb_dir=YOUR_PATH_TO_YOUR_TILEDB

You will need to provide the path to your tiledb.

PaSCient inference

We provide a notebook with examples on how to run PaSCient on your data in pascient/notebooks/paper_figures/model_inference_example.ipynb

Request model weights

Model weights are available upon request to debroue1@gene.com

Figures reproduction

Figures were generated using the notebooks in pascient/notebooks/paper_figures/

• UMAP.ipynb reproduces Figure 3 of the paper
• IG.ipynb reproduces Figure 4 of the paper
• severity.ipynb reproduces Figure 5 of the paper

Contact

If you have any questions, please contact Tianyu Liu (tianyu.liu@yale.edu) or Edward De Brouwer (debroue1@gene.com).