cantinilab/scc_reproducibility

scConfluence reproducibility

scConfluence is a novel method for the integration of unpaired multiomics data combining uncoupled autoencoders and Inverse Optimal Transport to learn low-dimensional cell embeddings. These embeddings can then be used for visualization and clustering, useful for discovering subpopulations of cells, and for imputation of features across modalities. This is the code used to perform the experiments and generate the figures in our manuscript. If you are looking for the Python package, click here!

Installation

install the scconfluence package from Pypi:

pip install scconfluence

Usage

Downloading the data

All the data used in the paper has been formatted and can be downloaded from here. All *.h5mu.gz files should then be placed in the data folder of this repository.

Obtaining each method's outputs

To obtain the outputs of scConfluence, run the following command:

python run_integration.py --dataname <dataname>

<dataname> is the name of the dataset which must be one of the following:

• cell_lines_ for k=0,1,2,3 where k is the number of the simulation scenario for the unbalanced celllines experiment
• pbmc10X
• OP_Multiome
• bmcite
• OP_Cite
• smFISH
• 3omics
• Patch

To obtain the outputs of the baseline methods, add the --baseline flag.

python run_integration.py --dataname <dataname> --baseline

Similarly, to obtain the imputation results on the smFISH dataset, run the following command:

python run_imputation.py --dataname <dataname> (--baseline)

Generating the figures

Before running any notebook, download the pickles from there and place them inside a folder entitled exp_results in the root of this directory. These files contain saved results (latent embeddings and their evaluations) of each method run with the provided configurations as the training of all methods is very long. Then run the notebooks to obtain the plots from the paper. As of now, only cell lines and benchmark plots are available. The rest will be added shortly.

Our preprint

https://www.biorxiv.org/content/10.1101/2024.02.26.582051v1

@article {Samaran2024unpaired,
  author = {Jules Samaran and Gabriel Peyre and Laura Cantini},
  title = {scConfluence : single-cell diagonal integration with regularized Inverse Optimal Transport on weakly connected features},
  year = {2024},
  doi = {10.1101/2024.02.26.582051},
  publisher = {Cold Spring Harbor Laboratory},
  journal = {bioRxiv}
}

scc_reproducibility's Issues

How to reproduce the figures

Hi! Recently I'm trying to reproduce the results of the paper, especially the baselines. But it seems that the output of run_imputation.py and run_integration.py is consists of a few folders named by algorithms (eg. run-scGLUE) which contains some csv files, and three pickle files: full_results.pickle, mean_results.pickle, median_results.pickle. However, to reproduce the final images, I need to run fig_bench.ipynb and fig_cline.ipynb, which need bench_rez.pickle and cline_rez.pickle as input. The structure and content of output results are simular to neither benchrez and clinerez, and not able to run the notebook. What's the correct procedure to reproduce the figures? Could you please give a more detailed instruction? Thanks.