MICrONS mouse visual cortex 1 mm³

The mouse watched Hollywood. Now it watches your chart.

A real cubic millimetre of cortex, cell by cell. 144,117 nuclei, 420 traced arbors, and 12,894 neurons replaying exactly what they did while the mouse watched the movies.

Loading the cube

Stimulus

00:00.0/ 01:00.0

Every recorded neuron, sorted by when it peaks. White line: population mean.

01 The piece

A cubic millimetre, taken apart and put back together.

In April 2025 the MICrONS consortium published the largest functional wiring diagram of a mammalian brain to date. A grain of sand of mouse visual cortex. They cut it into 27,972 slices, each 40 nanometres thick, fed it to five electron microscopes for about six months, and let machines and people trace every cell inside it.

This is that tissue. Not an artist's impression of a brain. Every point in the cube above is a nucleus the consortium detected. Every thread is a neuron they reconstructed. We only made it small enough for a phone.

The slab, laid on its side. Nuclei counted in 10 micron steps from the pia down. neurons inhibitory non neuronal You can see layer 1 almost empty, layer 4 packed, the white matter with almost no neurons. Counted in your browser from the same 144,117 points as the cube.

1.3 × 0.87 × 0.82 mm: the volume, measured in the living animal
200,000+: cells in the reconstruction
523 million: synapses found by automated detection
~75,000: neurons recorded while the mouse was awake
2 petabytes: of raw electron microscopy, at about 4 nm
12 days: of cutting, watched by humans in shifts, day and night

02 The screening

Before the knife, the cinema.

For a week, a 75 day old mouse stood on a treadmill at Baylor College of Medicine and watched 10 second clips while a two photon mesoscope read the calcium glow of its visual cortex. Fourteen scans. Four visual areas. Layers 2 to 5. Green light every time a neuron fired.

Then the mouse flew overnight to the Allen Institute in Seattle, the same piece of brain went under the electron microscope, and the recorded cells were matched to their reconstructed bodies by hand: 19,181 matches, 15,439 unique neurons. Function and wiring, in the same cells. That had never been done at this scale.

Now showing, per the paper

Mad Max: Fury Road 2015
Star Wars: The Force Awakens 2015
The Matrix 1999
The Matrix Reloaded 2003
The Matrix Revolutions 2003
Koyaanisqatsi 1982
Powaqqatsi 1988
Naqoyqatsi 2002

Plus Sports 1M clips, rendered first person scenes, and synthetic motion stimuli. The footage itself belongs to the studios and is not shown here.

03 Two modes

One cortex. Two things to watch.

Replay

Recorded data. Nothing generated.

12,894 coregistered neurons, each playing back its own measured response to the six clips every scan repeated ten times. Sixty seconds, 120 bins of half a second, averaged over the repeats. We interpolate between bins so it moves. That is the only thing we add.

When a thread lights up, that arbor belongs to the neuron that fired.

Sixty seconds of cortex. Every recorded neuron, stacked in rows and sorted by the moment it fires most. The diagonal is the sorting, not a discovery. The vertical streaks are real: moments in a clip when a crowd of cells fires together.

Market

Real tuning. Real synapses. Your chart as the stimulus.

Every swap draws a bar on the tape. Buys tilt it up, sells tilt it down, size sets the angle. Each neuron answers by its measured orientation preference, the same thing Hubel and Wiesel found with a slide projector in 1959. Then the activity crosses 8,128 proofread connections to the cells downstream.

Buys excite. Sells inhibit the whole sheet for a moment, then it recovers. It always recovers.

What each neuron is waiting for. Preferred orientation of all 12,894 recorded neurons, measured in the living mouse. More of them like horizontals and verticals than diagonals, a known bias of mouse visual cortex. A gentle green candle wakes the horizontal crowd. A god candle wakes the vertical one.

04 Token

$MOUSE

A memecoin on Robinhood Chain whose chart is wired into a real piece of cortex. It watched the movies. It will watch your entries too.

Contract coming soon

Launching soon

Chain id 4663. $MOUSE is a memecoin with no promise of anything. It is not affiliated with or endorsed by the MICrONS consortium, the Allen Institute, Baylor College of Medicine, Princeton University or IARPA. The science is theirs, public, and credited below.

05 Method and sources

What is real, what is reduced, what is ours.

Real

Every soma position and its neuron or non neuron call: the public nucleus detection and classification tables (171,815 detections, 89,062 called neurons). We show the 144,117 above 25 cubic microns, since smaller ones are mostly false positives.
Excitatory and inhibitory labels: the public soma classification model (61,926 and 8,031 cells).
420 neuron skeletons from the public v661 release, including the proofread neurons with axons. They are a sample: the full release has tens of thousands, and 360 of ours also have a recording, so their arbor lights up when they fire.
Responses, preferred orientation, selectivity, layer and area of 12,894 coregistered neurons, and 8,128 proofread connections (9,038 synapses), from the data release of Ding et al. 2025.

Reduced

The cube shows 144,117 of 171,815 detected nuclei, 12,894 of the roughly 75,000 recorded neurons (the ones matched to a cell body and released with the wiring paper), and a few thousand of the 523 million synapses. Nothing is resampled at random: each cut follows a published table.
Positions are quantized to 16 bits (about 25 nm of error). Skeletons are thinned to at most 260 segments per neuron, keeping every branch point and tip. We show skeletons, not the full meshes.
Responses are stored as 8 bit values scaled to each neuron's own peak, so every cell uses the full brightness range. Absolute amplitudes are not compared between cells.
Layer borders for L2/3, L4 and L5 come from the released labels. The L1 and L6 borders are read from the soma density profile and are estimates.
The release gives 120 values per neuron without a time axis. We read them as the 60 seconds of repeated clips in half second bins, the binning the paper describes. If that reading is off, the order of events is still real and the clock is not.

Ours

Market mode is a model. The mapping from a trade to a bar angle, the sweep of the bar across the slab, the rise and decay of activity, the synaptic gain and the inhibition after sells are our choices. The tuning and the wiring it runs on are measured.
The little screen in Replay is a film leader counting each ten second clip, not the footage. We do not know which frame the mouse was looking at in each bin, and the films are not ours to show.
Until the token trades, Market mode watches a rehearsal tape generated in your browser. After launch the server reads Swap events from the pool over the chain RPC and the tape is retired.
Access to the live annotation database (CAVE) needs an account, so everything here comes from the public buckets with no login.

The MICrONS Consortium. Functional connectomics spanning multiple areas of mouse visual cortex. Nature 640, 435 to 447 (2025). Read the paper
Ding, Z. et al. Functional connectomics reveals general wiring rule in mouse visual cortex. Nature 640, 459 to 469 (2025). Read the paper
MICrONS Explorer, cortical mm³ data portal. Allen Institute, Baylor College of Medicine, Princeton University. Open the portal
Public buckets, no login: the BossDB open data bucket on AWS and the iarpa_microns bucket on Google Cloud. Coordinate frame from the standard_transform package, minnie65 v1.4. Orientation follows the release notes: 0 degrees is a vertical bar, counted counter clockwise.
Hubel, D. H. and Wiesel, T. N. Receptive fields of single neurones in the cat's striate cortex. J. Physiol. 148, 574 to 591 (1959).