Fruit fly connectome simulation · Snake ’97

Watch a simulated network built from a fruit fly’s mapped neural connections play Snake. Replay recorded games, explore the 3D brain, and follow individual cells’ activity.

0.0 s / —Initial state
Loading verified recordings…Recorded replay

Snake

Initial board
SNAKE '9716 × 16
SCORE 000READY
Last action
↰ Left↑ Straight↱ Right

Structured sensors

No decision yet
Input scene contextFly-inspired coloursIllustrative paletteLFR

Select a sensor to locate its receiving neurons.

Brain activity

Loading anatomy…

Population colours · brightness shows recorded activity strength.

Before the first input
Drag to orbit · scroll to zoom · right-drag to pan— FPS

No cell selected · click a cell to explore it

Display settings Logarithmic · 4×
Display only · measured values unchanged

Log scale reveals small signals · zero stays quiet · numeric values are unchanged.

Highlight populations None selected

Select one or more populations

No highlights selected · select one or more labels to focus the brain and grid.

Recording details

Loading anatomy…

Loading samples…

Selected links: blue in · amber out · up to 64

Every simulated neuron

Population colours · contrast-adjusted activity
Outlined: no 3D coordinates
SELECTED CELL ·

Select a cell

MODEL & METHODS

How this simulation works

This viewer replays Snake games from a trained model built around a fruit fly’s connectome—a map of neurons and their connections. The activity shown here comes from a computer simulation, not a living fly.

From images to decisions

What the brain display shows

How well does it play?

The model uses real anatomical connections with simplified physiology. Including a cell in the simulation does not establish that it is necessary for the learned behaviour.

Reading the colours and controls

Select a population to highlight an annotated group of cells. In Activity change, Rising and Falling can narrow that selection. These controls change the display, not the recorded values. No individual cell is selected at startup. Click empty space in the brain view or use Clear selection to remove a cell’s marker and connection lines.

Find interesting cells ranks the same recording by average signal magnitude, variability (standard deviation), or the latest change. These measures describe activity, not a cell’s importance for playing the game. The optional history zoom enlarges small signals; its axis labels show the scale.

In structured-input recordings, the blue/green thumbnail is an illustrative scene, not a reconstruction of fruit-fly vision. The model receives the five displayed sensor values.

Recording method and research scope

Data sources and reproducibility

MaleCNS v1.0: FlyEM / HHMI Janelia, University of Cambridge, MRC Laboratory of Molecular Biology and Google Research, CC BY 4.0. We converted and simulated the released data; these modifications are not endorsed by its creators. Original data. Three.js license.