nd-embedding-atlas

Explore and annotate n-dimensional microscopy images with precomputed embeddings.

nd-embedding-atlas links precomputed embeddings to their source n-dimensional (TCZYX) microscopy images. Researchers can review, select, and annotate examples for downstream model training without uploading the dataset.

nd-embedding-atlas does not train models or include model-training code. Generate embeddings before launch and store them in a supported AnnData or MuData Zarr input.

Active development

APIs, file-format support, and shipped presets may change between releases. Check the supported formats before preparing production data.

Start here

Install ndea on macOS or Linux.
Check the browser and system requirements.
Launch ndea with an AnnData Zarr store or project YAML.
Confirm that your data matches the supported contracts.

ndea view path/to/data.zarr

Open http://localhost:5055 in Chrome or Edge.

Choose a path

Prepare data

Supported formats: compatibility boundaries for AnnData, MuData, OME-Zarr, and project YAML.
AnnData: required axes, observation columns, embeddings, and layers.
MuData: the supported axis=0 subset and modality rules.
Link observations to images: connect table rows to OME-Zarr fields of view.
Project YAML: mount several datasets, image plates, channels, and launch settings.
Optimise OME-Zarr imaging: choose pyramids, chunks, and sharding for interactive viewing.

Operate the application

CLI reference: commands, flags, environment variables, release channels, and exit codes.
WebGPU on HPC: enable hardware rendering in remote Chrome sessions.
Troubleshooting: diagnose launch, data, image, WebGPU, and update failures.

Contribute

Contributing: set up the monorepo and run focused checks.