χJump
not signed in

Assemble cells from fragments · · build 2026-09-04 18:29

Or paste x, y, z — or a whole Neuroglancer link — into the x field; it splits automatically.

Or browse a random cell

Good for learning what a cell type looks like — a new random example loads each click.

Recently viewed cells
Where it is in the block
The cells built so far

Click the fragments you want in the viewer — as many as you like — then copy the whole address bar and paste it here. Deselected fragments (the ones neuroglancer keeps with a !) are left out, and so is anything in a layer you have switched off.

The page sends you to a 40 µm column nobody has searched yet. The link opens with an empty annotation layer already selected, so ctrl-click every nucleus you can see, then copy the address bar and paste it below. Fifty nuclei in one paste. Everything already recorded in or near that column is drawn in grey, so you can see at a glance where not to mark again — including the ones just over the boundary, which are the ones you would otherwise record twice.

Every filter is applied in your browser over the cells this page holds — the ones proofread for the paper and the ones assembled here. Nothing is sent anywhere.

Select the cell(s) of interest

Leave everything unticked to match every kind. The counts are of cells that have a soma position, which is what every distance and layer test below needs.

Do you want to find cells close to another cell type?

Soma to soma, in three dimensions. cb2 is a 48 µm slab, so a radius much larger than that is measuring in x and y with the third axis nearly flat.

Which layer should the cell be in?

Has the cell been annotated with a specific organelle or extracellular structure?

Has the cell been annotated here in any other way?

What this tool records about a cell besides its organelles: who assembled it, what they called it, and whether the published table contradicts itself about it.

Should the cells fall inside specific bounding box(es)?
Add extra layers to the 3D view (optional — doesn’t change which cells match)

Every cell in cb2 this tool knows about: the ones proofread for the paper, the ones assembled here, and the nuclei somebody has marked with no cell built around them yet. Each row shows the cell as it stands now, so a reconstruction somebody has extended here says so and opens with the extra fragments in it. Add fragments aims the paste box at that cell, so you extend or correct an existing reconstruction rather than starting a rival one — an unassigned cell offers Build a cell here instead, because there is nothing to extend yet.

Nguyen, Thomas, Rhoades, Ricchi, Yuan, Sheridan, Hildebrand, Funke, Regehr & Lee, “Structured cerebellar connectivity supports resilient pattern separation”, Nature 613, 543–549 (2023), served by BossDB. Access, not a reuse licence: BossDB reports this data public and the paper names the repository, but no reuse terms are stated on the project page. Cite the paper, and check the terms before redistributing pixels.

Every other tool in this family classifies cells somebody else already found. cb2 has no such list — what it publishes is a segmentation of fragments, pieces of neurite split wherever the automatic segmentation was not confident. Putting them back together is the work, and it is what this page records.

A built cell carries the fragment ids and whatever people have called it. When you open one in the viewer its fragments are merged into a single object using neuroglancer's own equivalences, so anybody you send the link to sees one cell rather than forty pieces — without needing this page at all.

Everything here is counted from this page's own data when you open the tab — the proofread list embedded in the file, plus whatever the community backend has returned this session. Nothing is fetched to draw it.