Skip to content

A fuller cell, and one cell opened in the viewer - #71

Merged
ngodzik merged 5 commits into
mainfrom
feat/inside-a-cell
Nov 8, 2026
Merged

ngodzik merged 5 commits into
mainfrom
feat/inside-a-cell

Conversation

@ngodzik

@ngodzik ngodzik commented Nov 8, 2026

Copy link
Copy Markdown
Owner

The networks so far had seven proteins. This brings a fuller cell, eighteen proteins between what a cell reads and what it decides, and a window in the viewer that opens one cell to show them as they push (chapter 28).

What changes

  • A fuller cell. examples/networks/fuller-cell.toml fills in what the RAS/ERK network and the stress response leave out: EGFR, RAF and MEK before ERK, MEK and ERK each answering with a Hill response of order two for their double phosphorylation; cyclin D, RB, E2F and cyclin E, with p27 made by HIF-1α as the brake hypoxia puts on the cycle; MDM2, p21 and BAX around p53, BAX held back by BCL2. The cell stops dividing when RB is on. spheroid-fuller.toml and cord-stress-fuller.toml run it.
  • None of its rates is measured. They are set so that the fuller cell decides what the simpler networks decide, and crates/io/tests/fuller_cell.rs holds them there: ERK at 0.333 and the same target volume, RAS stuck on raising ERK to 0.553 against 0.556, RB crossing the arrest threshold at the oxygen where HIF-1α crosses it, BCL2 failing at the same anoxia, caspase within 0.015 of the stress response, one steady state however a cell starts.
  • A term that switches off. inhibited_by joins the catalogue of terms, the mirror of activated_by: - rate * source * self. Each term can now be read as it pushes, which the window draws.
  • One cell opened. In the viewer, a click on a cell opens a window beside the tissue that draws the cell's own network, laid out from the network itself: what the cell reads, each protein as a box in the colour the tissue takes when coloured by it, each term an arrow, green where it switches on, red where it switches off, as thick as it pushes now. Below, it says in words whether the cell divides and why, how large it asks to be, whether it is alive and its chance of apoptosis, and plots the deciding proteins. --open, --open-at, --follow, --protein and --window do the same from the command line, for recordings. The camera turns only while the window has the pointer and the focus.
  • Networks by name. cellweave_io::shipped_network("fuller-cell") reads a network shipped under examples/networks, with everything it extends, from copies compiled into cellweave-io, as the built-in RAS/ERK network already was. A program built on cellweave then carries the fuller cell without a checkout beside it, at a path fixed when it was built. Tests hold that every file under examples/networks is shipped and reads the same as from its file.
  • The course. Chapter 28 is the fuller cell and the window; what the model does not say becomes chapter 29. Chapter 2's catalogue gains inhibited_by, the glossary thirteen words, the README the window and the example.

What it does to a run

run result
the calibrated spheroid, fuller cell against the simpler networks the same by size: first arrest at a radius of 167.9 pixels in both, first death at 197.4 against 197.3, at 320 pixels a core of 166.7 against 166.6; thirteen steps ahead, because its first cell starts settled
the stressed cord of chapter 8 dies out with the fuller cell where it holds with the simpler one: no birth once the stress is on, one cell left at step 3000 against 24, through p21; without p21's brake it divides again

How it was checked

check result
format, lints, docs, the course's symbols, no data green
tests 223 green
oldest supported Rust (1.97), licences green
the viewer builds, formatted, no lint, licences checked

All of the above was replayed locally on this branch; the same-seed pairs are left to this pull request's CI.

🤖 Generated with Claude Code

ngodzik and others added 5 commits November 5, 2026 19:05
The camera turned with the mouse's motion while the left button was held,
and that motion is the device's, reported wherever the pointer is. A
press in the window let go outside it is never seen to end: from then on
every move of the mouse across the screen turned the camera, which is how
recordings left behind other windows came out turned half way through.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ey push

inhibited_by joins the catalogue of terms: - rate * source * self, the
mirror of activated_by, with the same optional Hill response. It is what
a kinase that inactivates its target does, or a protein that marks
another for destruction, and the catalogue had no way to say it.

Network::influences returns every term of a cell's own network as it
stands under given inputs: the node it moves, its kind, what it reads,
its rate, and what it adds to the node's rate of change now. Every
term's contribution is worked out in one place, term_rate, for the
derivatives and for this view alike; the built-in network still matches
the numbers captured before it was a file, bit for bit.

Simulation::inputs gives what a cell reads from where it sits, the
fields averaged over its pixels; the step builds its inputs through the
same function. DEATH_THRESHOLD, QUIESCENCE_THRESHOLD and DIVISION_RATIO
are public, so that a picture of a cell can say how far it is from each.
No run changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A click on a cell, one that does not drag, opens a window beside the
tissue. It draws the cell's own network, laid out from the network
itself: what the cell reads in the first column, each protein one column
past the furthest of what it reads, a link that closes a loop set aside
and drawn back under the boxes. Each protein is a box in the colour the
tissue takes when coloured by it; each term that reads something is an
arrow, green where it switches on, red where it switches off, dashed
where it acts by an absence, as thick as it pushes now. The proteins
that answer the simulation's questions say so underneath.

Below, the window says in words whether the cell divides and why, how
large it asks to be, whether it is alive and its chance of apoptosis,
and plots the deciding proteins since the cell was opened. Hovering over
a protein lists its terms as they push; clicking it colours the tissue
by it.

The open cell is outlined in white and ringed by a halo, and the camera
can follow it, keeping it in what the window leaves of the screen.
--open, --open-at, --follow, --protein and --window do from the command
line what the mouse does, so that a recording can open a cell.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ides

examples/networks/fuller-cell.toml fills in what the RAS/ERK network and
the stress response leave out, eighteen proteins where they have seven:
EGFR, RAF and MEK before ERK, MEK and ERK each answering with a Hill
response of order two for their double phosphorylation; cyclin D, RB,
E2F and cyclin E, with p27 made by HIF-1a as the brake hypoxia puts on
the cycle; MDM2, p21 and BAX around p53, BAX held back by BCL2. The cell
stops dividing when RB is on.

None of the rates is measured. They are set so that the fuller cell
decides what the simpler networks decide, and crates/io/tests/fuller_cell.rs
holds them there: ERK at 0.333 and the same target volume, RAS stuck on
raising ERK to 0.553 against 0.556, RB crossing the arrest threshold at
the oxygen where HIF-1a crosses it, BCL2 failing at the same anoxia,
caspase within 0.015 of the stress response, one steady state however a
cell starts, and initial levels at the well-fed steady state.

Run against the simpler networks: the calibrated spheroid is the same
by size (first arrest at a radius of 167.9 pixels in both, first death
at 197.4 against 197.3, at 320 pixels a core of 166.7 against 166.6),
thirteen steps ahead because its first cell starts settled. The stressed
cord of chapter 8 dies out with the fuller cell where it holds with the
simpler one: no birth once the stress is on, one cell left at step
3000 against 24, through p21; without p21's brake it divides again.

Chapter 28 is the fuller cell and the window that opens a cell; what the
model does not say becomes chapter 29. Chapter 2's catalogue gains inhibited_by,
the glossary thirteen words, the README the window and the example.
No other run changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A program built on cellweave that wanted the fuller cell had to read examples/networks/fuller-cell.toml from a cellweave checkout beside it, at a path fixed when it was built. cellweave_io::shipped_network("fuller-cell") now reads it, with the networks it extends, from copies of every file under examples/networks compiled into cellweave-io, as the built-in RAS/ERK network already was. Tests hold that every file there is shipped and that each reads the same as from its file. No run changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@ngodzik
ngodzik merged commit 4bf4376 into main Nov 8, 2026
8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant