Installation

cellrank requires Python ≥ 3.12. If you encounter any problems, feel free to open an issue.

pip / uv (recommended)

Install CellRank from PyPI:

pip install cellrank

Or, using uv, a fast, modern pip replacement:

uv pip install cellrank

Optional extras

CellRank provides optional dependency groups for common workflows:

pip install "cellrank[moscot]"  # moscot for optimal transport (RealTimeKernel)
pip install "cellrank[plot]"    # additional plotting dependencies

PETSc and SLEPc (faster large-scale solvers)

For large datasets, CellRank uses PETSc and SLEPc (via pyGPCCA) to compute macrostates and fate probabilities faster and with less memory than the default SciPy-based eigensolver. Install them together with CellRank via the petsc extra:

pip install "cellrank[petsc]"

petsc4py and slepc4py ship no pre-built wheels, so pip compiles PETSc and SLEPc from source. CellRank only needs a serial build, which requires a C compiler — no Fortran or MPI. On systems without a Fortran compiler or system MPI (e.g. a stock macOS or minimal Linux), pass configuration options so the build stays C-only:

PETSC_CONFIGURE_OPTIONS="--with-fc=0 --with-mpi=0 --with-debugging=0 --with-shared-libraries=1" \
    pip install "cellrank[petsc]"

Do not set SLEPC_CONFIGURE_OPTIONS — SLEPc inherits its configuration from the PETSc build. The build is a one-time compilation (a few minutes) and is then cached. If you would rather not compile from source, install PETSc and SLEPc via conda instead (see below). If you specifically need a distributed MPI build, install a system MPI and add mpi4py alongside the extra.

conda / mamba / pixi

CellRank is available on conda-forge. This installation method is convenient if you want pre-built PETSc and SLEPc — libraries for large-scale linear algebra that CellRank uses when computing macrostates or fate probabilities on large datasets — without compiling them from source (see the pip instructions above for the compile-from-source alternative):

mamba install -c conda-forge cellrank

Or, using pixi, a fast, modern conda replacement:

pixi add cellrank

To also install PETSc and SLEPc as pre-built conda packages (no compilation):

mamba install -c conda-forge cellrank petsc4py slepc4py

Note

Windows users: PETSc and SLEPc are not built for Windows on conda-forge (only Linux and macOS). Because the conda-forge cellrank package depends on pygpcca, which in turn requires petsc, mamba install -c conda-forge cellrank cannot be solved on Windows and fails with an error such as petsc [...] does not exist (perhaps a missing channel).

On Windows, install CellRank with pip install cellrank instead. PETSc and SLEPc are optional — without them CellRank falls back to a slower SciPy-based eigensolver for computing macrostates and fate probabilities, which is fine for small to medium datasets. To use PETSc and SLEPc on Windows, run CellRank inside WSL and install via conda there.

Development version

To install the latest development version from GitHub:

pip install git+https://github.com/theislab/cellrank.git@main

See Contributing guide for setting up a full development environment.