This repository implements the code for the NeurIPS 2024 paper GENOT: Entropic (Gromov) Wasserstein Flow Matching with Applications to Single-Cell Genomics
- GENOT provides neural entropic optimal transport estimators for
- linear OT (Wasserstein) and quadratic OT (Gromov-Wasserstein and Fused Gromov-Wasserstein)
- all of these in balanced and unbalanced formulations
- for any cost function
We demonstrate these capabilities on simulated data, where we have a closed-form solution, and a wide range of single-cell genomics tasks (see also moscot for (discrete) optimal transport applications in single-cell genomics).
Check out the notebooks directory for a wide range of examples, also covered in the paper.
You can install genot from source:
$ pip install .or in editable mode as
$ pip install -e .