Widgets¶

Every panel, tab and window, control by control. For what to do with them, see the how-to guides.

The main window¶

One window holds one or more views. A view is a scatter plot, a 3D viewer and their shared control panel — see working in several views.

Pane layout¶

Three buttons sit in the top-right corner of each pane:

Drag the splitter between the panes to change the ratio. View → Reset Layout restores the default stacked 50/50 arrangement.

Each pane also has a chevron toggle for its own sidebar.

The status bar¶

Bottom of the window. Shows a running selection count (Selected: 2,604) and is where the meta staleness banner appears when a project's snapshot is more than a day old.

The control panel¶

The sidebar on the left of the scatter pane. Toggle with C over the plot, or View → Toggle Figure Controls. Five tabs.

General¶

The column dropdowns have an inline filter box, which matters when your meta table has a hundred columns.

Labels only appear once you have zoomed in far enough that a manageable number of points is in view. They are then automatically arranged around their points to avoid covering points or each other — a label may shift to the other side of its point, and in very crowded spots labels that cannot be placed without overlap are hidden until you zoom in further. A short connector line ties each label to its point. Labels of selected points and of search matches get placed first. All of this is configurable via View → Labels: decluttering and the connector lines can be turned off, and labels that don't fit can be hidden (the default), shown dimmed (tucked behind the points), or shown normally. These choices are remembered across sessions (the Show Labels toggle itself is not — it resets to on, like the camera).

As an alternative to labeling every point, View → Labels → Declutter Mode → One Label Per Group shows a single label per unique value instead: the label sits at the edge of its group of points, kept within the current view, with connector lines fanning out to all of its points (drawn behind them). If a value's points form several clearly separated islands, each island gets its own copy of the label with its own local connector lines — rather than one label with lines running across the whole view. Labeling every unique value takes precedence: the extra island labels are only added where space is left, so they never crowd out another value's only label. Double-clicking a group label highlights/selects its points just like a regular label.

Embeddings¶

Recomputes the scatter plot's layout. See recompute the embedding for the full parameter reference.

Fidelity¶

Answers "is this embedding lying to me?".

Drawing KNN edges for All points on a large project will slow the plot noticeably. Selected only is usually what you want.

Cluster¶

See reclustering for the full method reference.

Settings¶

The 3D viewer¶

A Neuroglancer-style viewer showing the morphology of whatever is selected. Controls are in shortcuts. Press C over the viewer for its legend, and use its own sidebar toggle for viewer settings.

View → Synchronize Viewer (on by default) is what couples the scatter selection to the viewer. Untick it when working with very large selections — see the cost of a big selection.

Opening the data widgets¶

The Connectivity Table, Distance Heatmap and Feature Comparison read an embedding's features or distances, so what can be opened depends on the embedding rather than on the project. In a project with several embeddings the View entries are therefore submenus — View → Feature Comparison → connectivity (UMAP) — listing embeddings without the data as disabled. See the View menu.

A widget opened that way belongs to the embedding you picked: its title carries the embedding's name, each embedding gets its own instance, and the active embedding is left as it was. The keyboard shortcuts below always open the widget for the active embedding.

Connectivity Table¶

View → Connectivity Table (Shift+Cmd+C). Up- and downstream partners of the current selection, shown three ways:

• Table — the raw matrix, neurons as rows and partners as columns.
• Profile — one line per neuron across its partners, for comparing connectivity patterns at a glance.
• Network — a node-link diagram of the selection and its partners.

All three read from the same filtered view. The sidebar's Data tab shapes that view for every tab at once; the second sidebar tab holds display options and follows whichever tab is active.

Data controls¶

Always on top, Copy and Export CSV sit below the sidebar tabs and apply whatever tab you are on. Copy refuses views over 200 rows — use Export CSV for those.

Display controls¶

Node size tracks a neuron's total synapse count and edge width its connection strength. Very dense views are refused rather than drawn — lower the limits or raise the threshold when that happens.

Finding things in the scatter plot: double-click a row or column header in the table, or click a node in the network (or a point in the profile), and the scatter plot jumps to it.

Distance Heatmap¶

View → Distance Heatmap (Shift+Cmd+D). Pairwise distances within the selection, as a heatmap with rotated column headers.

Two modes: reading a precomputed matrix, or computing distances for the current selection from features. The second is capped at 5,000 selected neurons, with a hint shown when you exceed it. Disabled entirely for KNN-graph projects.

Feature Comparison¶

View → Feature Comparison (Shift+Cmd+F). Ranks the features separating two groups. See finding what separates two groups for the full control reference including scoring.

Click any table row for a per-feature detail dialog with its own distribution plot.

Find Nearest¶

Selection → Find Nearest…. For each query neuron, finds its top-N most similar neurons drawn from a candidate pool you define. Built for "select a group on one side of the brain, find the closest matches on the other side": the pool is independent of the Scope, so it can deliberately exclude the query neurons — something Scope cannot do, because scoping to the other side would deselect them.

The query set is pinned from the figure selection when the widget opens; it does not follow the selection afterwards unless you ask it to. Press Find Nearest Neighbours to fill the table — nothing is computed until you do, since the pool can be the whole dataset.

In Wide layout each neighbour cell shows the neighbour's id over its distance (label + full distance on hover); blank cells mean fewer than N in-pool neighbours were found (common with a KNN-graph source and a strict pool). Double-click a query or neighbour cell to jump to it in the scatter plot.

The action buttons sit below the table:

Always on top is on the left of the same row. The widget follows the active embedding: it refreshes on show, so reopen it after switching embeddings to pick up that embedding's similarity data.

The KNN source¶

A KNN-graph source stores only each neuron's top-k neighbours, so "nearest" is a lookup into that stored list, not a fresh search. A pool filter can only narrow the stored list — it can't reach a neuron that was never in the top-k. So with a strict pool you often get fewer than N results, or none. This matters most for left → right matching: a neuron's stored neighbours are usually same-side, so its contralateral match is rarely in the list. Prefer the feature or distance-matrix source there — those search the whole pool.

Search through graph (approximate) makes the KNN source reach further by walking the graph outward (shortest path over the stored edges) until it finds N in-pool members. Use it when a KNN graph is your only similarity source. The reported distances then become accumulated graph-path distances — an approximation, not the true query-to-neighbour distance — so the results are ranked by reachability rather than by exact similarity, and the distance column is labelled Graph distance to make that explicit.

Meta Data Explorer¶

View → Meta Data Explorer (Shift+Cmd+M). The full meta table, filterable, and the entry point to both meta-update routes.

The Update button is the only way into the two meta dialogs — they are not in the menu bar.

Meta Data Sources¶

Meta Data Explorer → Update → From Remote…. Configures where each dataset's annotations are read from, and pulls them. Never writes to a backend. See refreshing meta data.

Table columns: Dataset, Rows, Backend, Configuration, Column mapping.

In the mapping dialog, DO_NOT_UPDATE is the first choice for every column and means "leave this alone".

Update Meta From Local File¶

Meta Data Explorer → Update → From Local…. Merges a local file into the meta table. Accepts .csv, .tsv, .parquet and .feather. See route 2.

Meta rows never change; unmatched file rows are reported and ignored. Nothing is written to disk.

Push Annotations¶

Selection → Set Annotations (Cmd+A / Ctrl+A). The only dialog that writes to a live backend. See pushing annotations — read it before using this.

Build Project¶

File → Build Project… authors a new project directory from local tables, without hand-writing the info file. It is a form over the same engine as the ProjectBuilder Python API, so both produce identical projects.

Build project stays disabled until an output folder, a meta table and at least one embedding are set. It then writes the files; validation errors (e.g. a non-square distance matrix) are surfaced before anything is written. Leave Open the project after building (on by default) ticked to load the new project straight into the current view.

Command¶

Help → Command shows the command line that re-opens the current view — its source and filters — in both the uvx bigclust2@latest … and installed bigclust2 … forms, each with a copy button. Enabled once a project is loaded. See the command-line reference.

Other dialogs¶