Celcomen aims to fill an important gap in the literature:
Genetic screens in dissociated single cells → Virtual Cells
Genetic screens in spatial transcriptomics → ?
Celcomen is a causal generative model designed to disentangle intercellular and intracellular gene regulation with theoretical identifiability guarantees. Celcomen can then generate counterfactual spatial transcriptomic samples by simulating the effect of local perturbations.
Celcomen can
- predict the effect that a genetic perturbation on a cell will have on the cell and its neighbors,
- disentangle intra- and inter-cellular gene regulation,
- study differential gene regulation and cell communication between conditions, such between health and disease,
By enabling in-silico screening of perturbations, it can provide access to experimentally inaccessible samples, and accelerate scientific discovery.
You can find out more by reading our journal paper or ICLR publication.
We recommend using Anaconda/Miniconda to create a conda environment for using celcomen. You can create a python environment using the following command:
conda create -n celcomen_env python=3.9
Then, you can activate the environment using:
conda activate celcomen_env
Then install
pip install git+https://github.com/stathismegas/celcomen
To learn intracellular and extra-cellular gene regulation and then use it to simulate inflammation conuterfactuals in specific locaitons of the tissue, follow the tutorial analysis.spatial_KO.xenium_human_glioblastoma_gpu.ipynb.
As explained in the tutorial, the adata object should have count data, without any prior normalization or log-transformation.
More details about the documentation can be found on Read the Docs.
To reproduce our results from our paper and manuscript refer to our reproducibility repo.