Installation

scmidas requires Python ≥3.10 and is tested on Python 3.10 / 3.11 / 3.12 against PyTorch 2.5–2.10 and Lightning 2.4–2.6. We recommend creating a fresh environment to avoid conflicts with existing setups.

Step 1: Create a new environment

conda create -n scmidas python=3.12
conda activate scmidas

Step 2: Install scmidas

pip install scmidas

This pulls in the full set of dependencies (PyTorch, Lightning, scanpy, mudata, …). On Volta / Pascal GPUs (e.g. V100, P100, GTX 10xx), import scmidas emits a UserWarning with downgrade instructions if your CUDA install is incompatible with the resolved PyTorch wheel — newer torch (≥2.11) dropped those compute capabilities from its default cu128/cu129 wheels.

You’re all set. Continue with the Tutorials — the Quickstart gets you from pip install to a UMAP in about a minute.