concept · intermediate · partly written
How a multimodal single-cell or spatial data object ties its parts together
AnnData, MuData and SpatialData tie parts by a name or an integer id and by named coordinate systems; none stores how sure that tie is.
The tie in one paragraph
AnnData matches annotations to the matrix by a string index. MuData matches cells across modalities by cell name. SpatialData matches a table row to a mask pixel value or shape index by an integer id, and places images, masks and points in named coordinate systems. None of them checks that the match is true, and none stores an uncertainty for a mask or a transform. 123
Reported, not yet replicated
AnnData: a table and a string index more
X is cells by features, usually sparse. The tables obs and var annotate cells and features; obsm and varm hold arrays such as embeddings, layers hold alternative versions of X, and uns holds free metadata. Annotations align to X by the names in obs_names and var_names. Missing values have three encodings: -1 for categoricals, a boolean mask for nullable types, and NaN in floats. When objects are joined on an outer join, features missing from one object are filled with 0, so a feature that was never measured reads as a measured zero. 145
Reported, not yet replicated
MuData: several tables, one cell list more
MuData holds one AnnData per modality plus shared annotations, saved as .h5mu. The specification says that in practice it relies on obs_names and var_names to collate global observations and variables, and that an observation or variable missing from a modality corresponds to a 0 in the map. So an unmatched cell is a 0, not a row of NaN. The page is headed RFC, a request for comments, so it is not a ratified standard. 26
Established
OME-NGFF: the image layer more
OME-NGFF, with OME-Zarr as its storage, is a community specification for cloud-readable bioimaging data. Its RFC-5 defines named coordinate systems and transformations, and warns that coordinate systems with identical axes in the same order may not be the same. The RFC page read was at stage S3 (implementation), updated 2026-02-26. 78
Reported, not yet replicated
SpatialData: three kinds of tie more
The design document, marked superseded and read through a summarising tool, says there is no explicit link between elements: an image and a mask belong together only by sharing a coordinate system. Each element maps to at least one named coordinate system, with identity to global as the default; a transformation changes lightweight metadata, not the data. The supported transformations include identity, translation, scale, affine, rotation and sequence; MapIndex, Displacements, Coordinates, InverseOf and Bijection are listed as not supported, so non-linear warps are out in that version. A table links to an element through three keys, region, region_key and instance_key. For a mask the instance id must equal the mask pixel value; for shapes it is the index. A table without the keys is allowed but maps to no element. The newer specification page was not opened, so these limits may have changed. 39
Reported, not yet replicated
What the object does not hold more
A mask, a transform and a table row are each made by a vendor pipeline or a person, and the object records each outcome without a confidence. A measured error is therefore outside the format: see the record on segmentation and registration error. This reading is our, from the specifications, not a statement by the authors. 310
Our inference
Who defines what more
OME-NGFF is a defined community specification. AnnData, MuData and SpatialData are designs by single groups (the Scanpy, MUON and SpatialData authors) that became common practice; the authors of MuData argue that no open, language-neutral multi-omics exchange format existed, which is also the case for their own tool. Vendor formats from the imaging platforms were not opened. 5697
Reported, not yet replicated
Plain answer
Why it matters
Real numbers
Mass-spectrometry imaging dataset as SpatialData 2026-11-18
A mouse-brain MALDI dataset of 918,855 pixels was 18 GB raw and 4.83 GB as SpatialData (one group's tool; first author shared with SpatialData). 12
Reported, not yet replicated
Still open here
- Representing cells that are present in only some modalities (union with imputation or intersection) is an open issue in the MuData project, with no reply when read.
- The newer SpatialData specification was not opened.
- No micrometre registration error was found for Xenium-to-Visium or H&E-to-Xenium alignment.
Not yet written: Equation.
Linked from
- Imageromics (Amit's word) and the technologies it covers
- What the data object is, layer by layer
- Where the spatial tie is weakest: segmentation, registration and resolution
- How MOFA+, WNN and totalVI join two layers, and how they are benchmarked
- What a missing value means in each omics layer
- scverse
- How much of an imaging-based spatial molecule table is assigned to the wrong cell?
- What is the registration error, in micrometres, when spatial modalities are aligned?
Sources
- On-disk format (anndata documentation).. web, not peer reviewed; opened 2026-11-18 all records citing this
- MuData specification [RFC].. web, not peer reviewed; opened 2026-11-18 all records citing this
- Design document for SpatialData.. web, not peer reviewed; opened 2026-11-18 all records citing this
- scverse tutorial: concatenating AnnData and MuData objects.. web, not peer reviewed; opened 2026-11-18 all records citing this
- Wolf FA, Angerer P, Theis FJ. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol (2018). paper; opened 2026-11-18 all records citing this
- Bredikhin D, Kats I, Stegle O. MUON: multimodal omics analysis framework. Genome Biol (2022). paper; opened 2026-11-18 all records citing this
- Moore J, Allan C, Besson S, Burel JM, et al.. OME-NGFF: a next-generation file format for expanding bioimaging data-access strategies. Nat Methods (2021). paper; opened 2026-11-18 all records citing this
- RFC-5: Coordinate systems and transformations (OME-NGFF).. web, not peer reviewed; opened 2026-11-18 all records citing this
- Marconato L, Palla G, Yamauchi KA, Virshup I, et al.. SpatialData: an open and universal data framework for spatial omics. Nat Methods (2025). paper; opened 2026-11-18 all records citing this
- Yang Y, DePasquale E, Adeleke D, Xie X, et al.. MisTIC: Missegmented Transcript Inference Correction for Improved Spatial Transcriptomics Analysis. bioRxiv (2025). preprint, not peer reviewed; opened 2026-11-18 all records citing this
- Jones DC, Elz AE, Hadadianpour A, Ryu H, et al.. Cell simulation as cell segmentation. Nat Methods (2025). paper; opened 2026-11-18 all records citing this
- Visvikis T, Vierdag WM, Marconato L, Birmpili A, et al.. Native, Memory-Bounded Conversion of Mass Spectrometry Imaging Data into the SpatialData Ecosystem. J Am Soc Mass Spectrom (2026). paper; opened 2026-11-18 all records citing this
Where it sits on the map
History of this record (6)
This record has been changed 6 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.
- Status
- partly written
- Last updated
- 2026-11-19
- Record
- content/concepts/multimodal-and-spatial-data-objects.json