Tardigraph
Share and reuse

Resources

Everything we built is free to use, change and redistribute under CC BY 4.0. That means credit us and the original data. Units are micrometres. Axes are +x anterior (head), +y dorsal (back) and −z the animal's left.

Cite both this project and the data it's built on; see how to cite. The organ names and the placement of the muscles are our identifications; the pages on the X-ray and muscles say how each was made and what's uncertain.

3D models

FileWhat it isUse it in
tardigrade-organs.glb 11.5 MBBody surface and 19 organs and structures, one named mesh eachWeb (three.js, model-viewer), Blender, Unity, Unreal
tardigrade-organs-stl.zip 13.8 MBThe same, one STL per structure3D printing, CAD
tardigrade-organs-obj.zip 7.3 MBThe same, one OBJ per structureBlender, MeshLab, most 3D tools
tardigrade-muscles.glb 7.2 MB100 leg muscles (M1–M14, legs I–IV, both sides), placed on the X-ray animal, named like iiM5-LSame as the organs; fits them in the same frame
tardigrade-muscles-stl.zip 8.8 MBThe same, one STL per muscle3D printing
How this is built
Data
The nano-CT scans and organ label fields of Gross et al. 2019 (CC BY 4.0), and the confocal leg-muscle label fields of Gross & Mayer 2019 (Dryad, CC0).
Built with
build_meshes.py turns each organ label into a surface (marching cubes, then simplified) in micrometres. build_muscles.py names each muscle from the paper's colour code and fits each leg onto the CT animal by two landmarks, foot and dorsal attachment, then nudges strands out of the organs. export_resources.py writes the GLB, STL and OBJ files.
Checked
By hand: the label files carry no names, so we matched each organ's measured volume to the paper's Table 1 and checked its position and number of pieces; uncertain ones are marked "probable". The muscle colours in each leg match Table 1's counts exactly (14, 12, 11 and 10).
Limits
The muscles come from a different animal and are placed approximately. Their slice spacing is estimated from round muscle strands, an assumption, and the right side is a mirror of the left.
More
How the whole project is built

Volumes and tables

FileWhat it isUse it in
tardigrade-labels-270nm.nrrd 0.5 MBOne label volume for the whole animal at 270 nm. It combines the published whole-body labels with the brain, salivary glands and head structures from the head scan, registered in, plus the traced body. Not available anywhere else.3D Slicer, napari, ITK-SNAP, Python (pynrrd)
tardigrade-labels-270nm.tif 0.6 MBThe same as a multi-page TIFFFiji / ImageJ
tardigrade-structures.csvLabel ids, names, confidence, measured and published volumes, and how each was identifiedAnything
tardigrade-ct-bodymasked.nrrd 5.9 MBThe whole-body CT (8-bit), everything outside the traced body set to zeroVolume rendering, segmentation
tardigrade-gait.jsonEvery step of all eight legs in one recorded walk, group gait means (adults, juveniles, Ramazzottius, disulfiram), and the leg geometry on the X-ray animal. From Anderson et al. 2024.Animation, robotics, gait analysis
How this is built
Data
Gross et al. 2019's whole-body scan and labels at 270 nm and their head scan's labels; walking data from Anderson et al. 2024's repository (no licence stated).
Built with
export_resources.py merges the whole-body labels, our traced body and the head-scan labels. We placed the head scan by cross-correlating the two scans over all 24 orientations, then refined the shift on the pharynx, buccal tube and ventral ganglion. export_gait.py turns the walking workbooks into step events and group means.
Checked
The head scan's stated voxel size did not fit its data. At the size we measured, 0.1065 µm, its brain and salivary glands come to 1,389 and 2,274 µm³, against the published 1,393 and 2,274.
Limits
The head-scan voxel size is our measurement, not the paper's figure. The walk is one recorded animal plus group means.
More
How the whole project is built

Predicted protein structures

Eleven co-folding predictions from the condensin II work: each tardigrade CAP-H2 kleisin region placed on its own species' SMC head, with the human pair as a positive control. Each file is the best-scoring sample of several.

These are computed models, not experimental structures. ipTM is the confidence in the interface between the two chains and pLDDT the per-residue confidence of the kleisin; scrambled-sequence controls scored 0.19–0.59, so treat anything below about 0.6 as unsupported. They cannot be deposited in the PDB, which takes experimental structures only — ModelArchive is where computed models belong and is where these are headed.

FileWhat it isConfidence
c_pme_a.pdb 435 kBP. metropolitanus — Paramacrobiotus metropolitanus CAP-H2 copy a, C-terminal region, on its own SMC4 headipTM 0.808, pLDDT 62
c_pme_b.pdb 426 kBP. metropolitanus — Paramacrobiotus metropolitanus CAP-H2 copy b, C-terminal region, on its own SMC4 headipTM 0.805, pLDDT 55
c_hex_b.pdb 408 kBH. exemplaris — Hypsibius exemplaris CAP-H2 copy b, C-terminal region, on its own SMC4 headipTM 0.794, pLDDT 66
c_hex_a.pdb 412 kBH. exemplaris — Hypsibius exemplaris CAP-H2 copy a, C-terminal region, on its own SMC4 headipTM 0.777, pLDDT 50
c_rva_b.pdb 413 kBR. varieornatus — Ramazzottius varieornatus CAP-H2 copy b, C-terminal region, on its own SMC4 headipTM 0.763, pLDDT 63
c_rva_a.pdb 430 kBR. varieornatus — Ramazzottius varieornatus CAP-H2 copy a, C-terminal region, on its own SMC4 headipTM 0.677, pLDDT 38
n2_hex_b.pdb 452 kBH. exemplaris — Hypsibius exemplaris CAP-H2 copy b, N-terminal region, on its own SMC2 head and neckipTM 0.665, pLDDT 68
n2_rva_b.pdb 451 kBR. varieornatus — Ramazzottius varieornatus CAP-H2 copy b, N-terminal region, on its own SMC2 head and neckipTM 0.655, pLDDT 64
n2_pme_b.pdb 443 kBP. metropolitanus — Paramacrobiotus metropolitanus CAP-H2 copy b, N-terminal region, on its own SMC2 head and neckipTM 0.415, pLDDT 44
c_hsa.pdb 400 kB controlHomo sapiens — Human CAP-H2 C-terminal region on human SMC4 head, as the positive controlipTM 0.834, pLDDT 74
n2_hsa.pdb 451 kB controlHomo sapiens — Human CAP-H2 N-terminal region on human SMC2 head and neck, as the positive controlipTM 0.613, pLDDT 58

Method and the full sample set: the condensin II preprint, doi:10.5281/zenodo.22832510. Predicted with Boltz-2.

Minecraft

FileWhat it isUse it in
tardigrade_270nm.schem 0.3 MBGlass skin and concrete organs, 608 × 188 × 213 blocksJava 1.20+ with WorldEdit, FAWE or Litematica
tardigrade_ct_270nm.schem 2.1 MBThe X-ray itself in 11 grey stones, solidSame
README.txtHow to load them, block legend, credits
How this is built
Data
The same nano-CT and labels (Gross et al. 2019, CC BY 4.0), with the head structures placed by our registration.
Built with
export_minecraft.py turns one 270 nm voxel into one block and writes Sponge schematics: a one-block glass skin with concrete organs, and a second build where each voxel's CT brightness picks one of 11 grey blocks.
Checked
Not tested against data; it is a direct conversion of the scan.
Limits
One block stands for 270 nm, so the animal is about 3.7 million times life size and carries the scan's resolution, no finer.
More
How the whole project is built

The knowledge graph

Everything the project knows, as 2,500 linked facts, sources, open questions, tests and assumptions, each citing the page or record it came from. Search it in the browser, or download nodes.jsonl.gz and edges.jsonl.gz. CC BY 4.0.

Videos

Walking, the block build and the stone build, each wide, square and tall: on the front page. Share freely with credit.

Tools

We wrote these along the way. Each works beyond tardigrades. The code is on GitHub: tardigraph-tools (MIT licence), and the condensin analysis has its own repository.

ToolWhat it does
Label fields to meshesTurns Amira and ImageJ label volumes into named, simplified meshes (GLB, STL, OBJ). Traces a body surface from a CT cuticle, handling mounting film.
Scan-to-scan registrationPlaces a high-resolution partial scan inside a whole-body scan: FFT cross-correlation over all 24 orientations, then refinement on label overlap.
Muscle placementFits segmented structures from one specimen onto another by landmarks, then nudges them out of the target's organs.
Voxels to MinecraftWrites Sponge schematics from any label or greyscale volume, with no dependencies beyond NumPy.
Browser viewerA three.js viewer for organ meshes: glass body, CT on the cut face, grouped structure list, walking from step events, shareable view links.
Video recorderRenders frame-exact videos of browser 3D scenes in wide, square and tall, on the GPU.
Reciprocal ortholog checkTests whether "missing" genes are really missing: reciprocal best hits and PSI-BLAST across related proteomes.
Remote tar indexingFinds files inside huge remote tar archives (such as OpenGenome2) with HTTP range requests, without downloading them.
Methylation from HiFi BAMsReads 5mC calls (MM/ML tags) from public PacBio files, with checks for amplified libraries.

Preprints

Everything we have written up, with its data and the script that produced it. All CC BY 4.0, all open, all with their failures and corrections in the text rather than trimmed out of it.

How to cite

Please cite this project, and the data you used.

Meow-Ludo (2026). Tardigraph: a water bear rebuilt from published data. https://tardigraph.meow-meow.io/ (accessed [date]). CC BY 4.0.
@misc{tardigraph2026,
  author = {Meow-Ludo},
  title  = {Tardigraph: a water bear rebuilt from published data},
  year   = {2026},
  url    = {https://tardigraph.meow-meow.io/},
  note   = {CC BY 4.0}
}

And the original data:

Full list: credits.