Installation
AutoZyme ships as two packages: autozyme (Python, on PyPI) and autozyme (R, from the GitHub repository; CRAN release pending). Both have no hard dependencies on any upstream package; each patch activates only when its upstream is importable.
Python
pip install autozyme
Requires Python 3.10 or newer. Prebuilt wheels cover Linux (x86_64 + aarch64, manylinux + musllinux), macOS (arm64 + x86_64), and Windows (x86_64) for CPython 3.10–3.14, so there is nothing to compile. The wheel has no runtime dependencies (no scanpy, no torch, no tensorflow).
Optional: install upstream packages
autozyme is useless without something to accelerate. Install whichever upstream(s) you actually use:
pip install scanpy # for autozyme.activate("scanpy")
pip install cell2location # for autozyme.activate("cell2location")
pip install scvelo # for autozyme.activate("scvelo")
pip install MDAnalysis # for autozyme.activate("mdanalysis_rmsd")
# ...
See the patch catalog for the full list and which upstream version each patch is pinned against.
Confirm the install
import autozyme
autozyme.list_patches() # all 16 patches discovered
autozyme.list_patches(installed=True) # only those whose upstream is importable
Or from the shell:
python -m autozyme
prints a one-screen dashboard of every patch and its upstream availability.
R
# from GitHub (CRAN release pending); the R package lives in the autozyme_r/ subdir
remotes::install_github("ElliotXie/autozyme", subdir = "autozyme_r")
# once published to CRAN
install.packages("autozyme")
The R package compiles C++ (Rcpp) kernels at install. This only really trips up Windows (no built-in compiler): install Rtools matching your R version first, or the build fails. macOS (Xcode Command Line Tools) and Linux (gcc) normally already have a compiler.
Requires R 4.0 or newer. Depends: is empty — Suggests: lists every upstream the lifted patches recognize, but installing autozyme doesn’t pull any of them in.
Confirm the install
library(autozyme)
list_patches() # all shipped patches
list_patches(installed = TRUE) # only those whose upstream is installed
dashboard() # one-screen status of patches + upstream availability
Platform support
Python ships prebuilt wheels (nothing to compile). The R package compiles from source, which only needs setup on Windows (Rtools).
| Platform | Python (prebuilt wheels) | R |
|---|---|---|
| Linux x86_64 / aarch64 (manylinux + musllinux) | ✓ 3.10–3.14 | ✓ |
| macOS arm64 / x86_64 | ✓ 3.10–3.14 | ✓ |
| Windows x86_64 | ✓ 3.10–3.14 | ✓ (needs Rtools) |
All patches are tested across Linux + Windows in CI; macOS is tested by maintainers but not on every PR.
Updating
pip install --upgrade autozyme
The R package installs from GitHub (CRAN pending), so re-run the install to pull the latest:
remotes::install_github("ElliotXie/autozyme", subdir = "autozyme_r", force = TRUE)
Releases follow PEP 440 / semver. Breaking API changes (e.g. a register_patch kwarg rename) bump the minor version and are gated by the API signature snapshot test — you’ll see the change in the changelog before it can land.
Uninstall
pip uninstall autozyme
remove.packages("autozyme")
autozyme makes no system-wide changes outside its own package directory. Removing it leaves your scanpy / Seurat / etc. installs untouched.