cantinilab/ot-scomics

This Python package will allow you to replicate the experiments from our research on applying Optimal Transport as a similarity metric in between single-cell omics data.

Optimal Transport improves cell-cell similarity inference in single-cell omics data

This Python package will allow you to replicate the experiments from our research on applying Optimal Transport as a similarity metric in between single-cell omics data.

Optimal Transport for single-cell omics

We propose the use of Optimal Transport (OT) as a cell-cell similarity metric for single-cell omics data. The code in this repository implements entropic-regularized OT distance computation with PyTorch, and applies it to public datasets of single-cell omics data. We compare the results to commonly used metrics like the euclidean distance or Pearson correlation, and demonstrate that OT increases performances in cell-cell similarity inference.

How many cells can I use?

Computing pairwise distance matrices is expensive (O(n²) operations for n cells), so we advise users of the package to limit the number of cells studied to a few thousand.

Installing the package

pip install otscomics

Documentation

https://ot-scomics.rtfd.io

Jupyter Notebook

The documentation includes a Jupyter Notebook demonstrating the package. While this notebook can be run un CPU, computations will be faster on GPU.

If you do not have access to a GPU you may want to use the (free) Google Colab platform to run this notebook: colab.research.google.com/github/ComputationalSystemsBiology/.../OT_scOmics.ipynb.

Please make sure that the GPU is enabled. Navigate to "Edit→Notebook Settings" and select "GPU" from the "Hardware Accelerator" drop-down.

Citing us

The paper describing OT-scOmics has been published on Bioinformatics.

Open Access link (Huizing, Peyré, Cantini, 2022)

@article{10.1093/bioinformatics/btac084,
	author = {Huizing, Geert-Jan and Peyré, Gabriel and Cantini, Laura},
	title = "{Optimal transport improves cell–cell similarity inference in single-cell omics data}",
	journal = {Bioinformatics},
	volume = {38},
	number = {8},
	pages = {2169-2177},
	year = {2022},
	month = {02},
	issn = {1367-4803},
	doi = {10.1093/bioinformatics/btac084},
	url = {https://doi.org/10.1093/bioinformatics/btac084},
	eprint = {https://academic.oup.com/bioinformatics/article-pdf/38/8/2169/43370259/btac084.pdf},
}

ot-scomics's Issues

Pip installable

Hi! We've been using your package as part of a benchmarking pipeline, however, when packaging our pipeline it is non-trivial to include a non-pip installable package like otscomics (currently we have a not so nice workaround so would be great to be able to just pip install otscomics).
The process is: 1) sign up to pip: https://pypi.org/account/register/, 2) in your OT-scOmics directory run: python setup.py sdist bdist_wheel, 3) check your built package (e.g. create a fresh env and try running your notebooks), 4) use twine (install via https://pypi.org/project/twine/) to check the package: twine check dist/* finally, 5) upload to pip: twine upload dist/*. Perhaps a thought would be to provide two builds (with and without cuda enabled) to make it accessible to different envs. Let me know if you have issues and please let me know when it becomes available, cheers!