Skip to content

How selection works

Selection is the central verb. Almost everything you do in BigClust is: select some points, then ask a question about them. Understanding what selection propagates to — and what it costs — is most of understanding the app.

One selection, every view

A selection made in the scatter plot is immediately the subject of everything else in that view:

  • The 3D viewer loads the morphology of the selected neurons.
  • The connectivity table shows their partners.
  • The distance heatmap shows distances among them.
  • The feature comparison takes them as a group.
  • Set Annotations writes to exactly them.
  • Copy to Clipboard copies exactly them.

There is no separate "apply" step and no per-widget selection to keep in sync. This is what makes the app feel like a single instrument rather than a collection of panels: you point at something once.

The selection is per view. Two views of the same project have independent selections, which is what makes comparing two things possible.

Making a selection

Gesture Effect
Shift + drag Box selection
Shift+Ctrl + drag Freehand lasso
add Cmd to either Add to the current selection instead of replacing it
double-click a label Highlight points sharing that label
Shift + double-click a label Select points sharing that label
Cmd+Shift + double-click a label Add those points to the selection
Cmd+A Select all
Cmd+I Invert
Esc Deselect all

The label gestures are worth internalising. If your meta table already has cell types, shift + double-clicking a type name selects every member of it — which is the fastest possible way to ask "where does this known type sit in my embedding, and is it in one place?".

Add as group in the General tab (on by default) means each new addition is tracked as a distinct group rather than merged into an undifferentiated set.

Growing and shrinking

Cmd++ grows the selection, Cmd+- shrinks it back.

Growing pulls in the nearest points to what is already selected — in the original high-dimensional space, not in the 2D plot. This distinction matters. Growing does not simply take the points that look nearby on screen; it takes the neurons that are actually most similar, which may include points the embedding has placed elsewhere. If growing repeatedly pulls in something from across the plot, that is your embedding telling on itself.

Selection → Grow/Shrink Options controls how:

  • Which data source the neighbourhood is computed from, and with which feature metric.
  • Grow By — a fixed number of points per step, or everything within a neighbour distance, optionally scaled by a Distance × factor.

Shrinking undoes the last grow rather than eroding the boundary, so grow-grow-shrink returns you exactly to where one grow left you.

This is the intended way to work outward from a seed. Select one neuron you are sure about, grow until the additions stop making sense, and you have delineated a type by hand with the data doing the work.

Hiding rather than deselecting

Three different operations that look similar:

Operation Effect Reversible
Esc Clears the selection; points stay
H (Hide Selection) Points stay loaded but are not drawn Alt+H shows all hidden
Backspace (Remove from View) Points leave the view entirely No — reopen the view
Scope filters (General tab) Points hidden by a rule, not by hand Yes, edit or remove the filter

Hiding is how you work through a large dataset: deal with a cluster, hide it, and what remains is what you have not looked at yet. Scope filters do the same thing declaratively — "hide everything that already has a type" is a filter, not a gesture.

Remove from View only makes sense in a selection view, which is why it is bound to Backspace there.

The cost of a big selection

Selection is cheap. Its consequences are not.

Every selected neuron is a mesh the 3D viewer has to fetch from a Neuroglancer source and render. Selecting 50,000 points asks for 50,000 meshes. BigClust confirms before very large selections for this reason.

Two ways to work with big selections comfortably:

  • View → Synchronize Viewer — untick it and the scatter selection stops driving the 3D viewer. Selection becomes instant. Use this whenever the question is about which neurons rather than what they look like.
  • The Settings tab has a neuron cache with a size limit and a Clear cache button. Caching helps a lot when you keep returning to the same groups.

Why plain letters are shortcuts

H, L, C, Shift+C and Backspace are bound as bare keys, which would normally be hostile — they would hijack typing in a search box.

They are scoped to the canvas widget rather than the whole window, so they only fire when the scatter plot has focus. Typing hello into the Meta Data Explorer's filter field does not hide your selection. The menu entries exist so the shortcuts are discoverable; the canvas is where they actually fire.


That is the last concepts page. For the full inventory of controls, see the reference.