======================================== treeVAE-Reproducibility

Reproducing results in the "Reconstructing unobserved cellular states from paired single-cell lineage tracing and transcriptomics data" paper, accepted at the ICML 2021 Workshop on Computational Biology.

Contact

ouardini.k@gmail.com

Datasets

1. Gaussian Process Factor Analysis (GPFA)

   • The simulated tree topologies used in the GPFA experiments are stored in scvi/data/topologies/100cells.
   • The code for the simulations is in scvi/dataset/ppca.py.

2. Gaussian Process Poisson Log Normal (GPPLN)

   • The simulated tree topologies used in the GPPLN experiments are stored in scvi/data/topologies/500cells.
   • The code for the simulations is in scvi/dataset/poisson_glm.py.

3. Metastasis

   • The tree toplogy fot the cancer metastasis dataset is stored in scvi/data/metastasis/lg7_tree_hybrid_priors.alleleThresh.processed.ultrametric.annotated.tree
   • The gene expression data (603 cells, 100 genes) is stored in scvi/data/metastasis/Metastasis_lg7_100g.npy .

System requirements

+ Python 3
+ Pytorch

Installation guide

+ Clone the github repository, create an environment and install the dependencies: `pip install -r requirements.txt`.

Instructions to reproduce experiments

+ for the GPFA experiments, run `python3 gaussian_ancestral_imputation.py`
+ for the GPPLN experiments, run `python3 ancestral_imputation.py`
+ for the metastasis data analysis, symply follow instructions in the notebook `scvi/external/notebooks/Metastasis.ipynb`

At the end of each run, the raw results will be stored in csv format in a results/ folder, with sub-directories automatically created in reference to the hyperparameters used in each experiment. to reproduce the tables and figures of the paper, simply follow the instructions in scvi/external/notebooks/plot_results.ipynb.