Adding initial version of CellBender WILDS Docker image

[tefirman]

Type of Change

• New Docker image

Description

Adds a Docker image for CellBender v0.3.2, a tool for removing ambient RNA and barcode swapping artifacts from droplet-based scRNA-seq and snRNA-seq count matrices using a deep generative model (published in Nature Methods 2023).

• Base image: python:3.11-slim
• Installation method: pip (from GitHub main branch, see note below)
• Key dependencies: PyTorch (CUDA-enabled for optional GPU acceleration), PyTables/HDF5 for .h5 file I/O
• Compatible with both linux/amd64 and linux/arm64

Note on version pinning exception: CellBender is installed directly from the upstream main branch rather than a pinned PyPI release. The v0.3.2 PyPI release has a torch.save checkpoint serialization bug (broadinstitute/CellBender#444) that causes runs to fail in practice. The fix is merged on main but not yet in a release. This image will be updated to pin a specific version once v0.3.3 is published.

Testing

How did you test these changes?

Ran make lint IMAGE=cellbender with hadolint v2.12.0 and validated end-to-end by running a WDL workflow against the built image.

Did the tests pass?

Yes — lint passed clean and the WDL workflow completed successfully.

Checklist

• Dockerfile follows naming convention (Dockerfile_X.Y.Z or Dockerfile_latest)
• All required OCI metadata labels are present and accurate
• README.md is included/updated in the tool directory
• Tested locally with make validate IMAGE=toolname (or manually built and verified)
• Image builds successfully for target platform(s)

Additional Context

PyTorch is installed with CUDA support (the default pip wheel) so users can take advantage of GPU acceleration via --cuda on GPU-enabled machines. CellBender falls back to CPU automatically when no GPU is detected.