Supported formats
Implemented AnnData, MuData, OME-Zarr, and project YAML contracts.
nd-embedding-atlas implements the subsets below. A file can be valid according to an upstream specification and still use an encoding or layout that ndea does not read.
| Input | Detection | Implemented subset | Important limits |
|---|---|---|---|
| AnnData Zarr | root encoding-type: anndata | Zarr v2 and v3 stores; obs, optional var, dense obsm, layers, categorical and string columns | obs is required; sparse embeddings are not supported |
| MuData Zarr | root encoding-type: MuData | MuData 0.1.0-style stores with axis: 0; AnnData modalities under mod/ | only axis=0; modality observations must map one-to-one to the shared observations |
| OME-Zarr HCS | plate metadata plus image multiscales | OME-NGFF 0.4 on Zarr v2 and OME-NGFF 0.5 on Zarr v3 | mounted plates only; numeric pyramid dataset paths and standard plate/well/image hierarchy are expected |
| Project YAML | .yaml or .yml launch path | array or mapping dataset syntax, relative paths, plate mounts, channel display settings, server settings | exactly one YAML file per launch; malformed or conflicting aliases stop startup |
AnnData element encodings
The reader handles dataframe columns represented as plain arrays, nullable arrays, string arrays, and categoricals. It reads dense arrays and CSR/CSC matrices where the relevant code path supports matrix data. Embeddings in obsm must resolve to dense two-dimensional arrays.
The implementation recognises AnnData root encoding 0.1.0 and dataframe/array-family encodings used by current AnnData Zarr writers, including dataframe and string-array 0.2.0 metadata.
Storage locations
Local filesystem paths are the supported launch interface. The lower-level Zarr reader can open HTTP(S) stores, but the application CLI, project validation, and mounted imaging pipeline are designed and tested around local paths.
Before converting data
- Open the store with
ndea viewand read the first diagnostic. - Confirm the root
encoding-typeand Zarr generation. - Check the focused format page for required groups and column names.
- Convert only the unsupported element rather than rewriting the entire store.
See AnnData, MuData, Image linkage, and Optimise OME-Zarr imaging for preparation details.