Connectomes are being run, embodied, turned into games and put to the test all over the world. This page maps the projects closest to Connectome Lab, what each one does differently, and what we can learn from them. Collected in September 2026; every entry links to its source.
Research groups that simulate a mapped nervous system, sometimes with a body.
Philip Shiu, Kristin Scott lab (Berkeley)
Leaky integrate and fire model of all FlyWire neurons; predicted feeding and grooming circuits (Nature 2024).
Compared with this lab: The model Connectome Lab runs. The original is Python and Brian2 on a desktop, without a body or tasks.
Eon Systems (San Francisco)
The Shiu brain model with the flyvis visual system and the NeuroMechFly body: grooming, feeding, foraging and a partial escape.
Compared with this lab: The closest concept, aimed at brain uploading. It runs on servers, is shown as video, and publishes no rewired or random controls. Carboncopies has argued that its body routines hide brain errors.
Turaga lab (Janelia), Macke lab
Connectome constrained network of the fly motion pathway whose weights are trained; predicts responses of 64 cell types (Nature 2024).
Compared with this lab: Trained weights on real topology, vision only. Connectome Lab keeps synapse counts fixed.
Janelia and Google DeepMind
MuJoCo fly body for walking and flight, controlled by trained neural networks (Nature 2025).
Compared with this lab: Connectome Lab's fly body. Its original controllers are trained networks, not connectomes.
Ramdya lab (EPFL)
Fly body with 87 joints, vision, smell and navigation, and a Python gym interface (Nature Methods 2024).
Compared with this lab: The other body standard; Python, not the browser.
Tsinghua University
The whole brain connectome as the graph of a reinforcement learning policy for flybody locomotion.
Compared with this lab: Trains weights on the real topology; unclear whether shuffled connectomes were tested.
BAAI and Peking University
Biophysical C. elegans network with a soft body in fluid; chemotaxis and perturbation experiments (Nature Computational Science 2024).
Compared with this lab: The most detailed whole worm today, on GPUs rather than in a browser.
Open community since 2011
The original open whole worm simulation: NeuroML network and fluid body.
Compared with this lab: The historic peer; slow to run and without a task benchmark. Recent activity not checked.
Ölveczky lab (Harvard) and DeepMind
A network trained to move a MuJoCo rat like a real one predicts striatum and motor cortex activity (Nature 2024).
Compared with this lab: No connectome: the rodent analogue of flybody's trained controllers.
Sandia National Laboratories
The full FlyWire connectome on 12 neuromorphic chips.
Compared with this lab: A hardware port without behaviour tasks.
The small movement that tests real wiring against shuffled or random wiring, like this lab does.
Brandon Cho
Pre-registered reflex benchmark with 36 tasks on FlyWire and MaleCNS, shuffled wiring controls, an embodied track and an open leaderboard.
Compared with this lab: Controls like ours, reflex tasks in Python; its leaderboard accepts other simulators, so this lab could submit.
Recluse
Compares the full connectome, a two line rule, replayed output and a degree preserving shuffle on NeuroMechFly.
Compared with this lab: Found the connectome equal to a simple rule on steering: a useful null result and a control we do not yet run.
OPC Studio
A 3,963 neuron FlyWire pursuit circuit aims and plays Doom without training; real wiring 10.2 kills per episode against 0.7 for scrambled wiring.
Compared with this lab: The closest in spirit: untrained circuit, controls and lesions, one game.
eganeganegan
MaleCNS against random, rewired and standard neural networks on VizDoom.
Compared with this lab: Controls on a game; no advantage claimed yet.
opifor
Pre-registered FlyWire, BANC and MaleCNS experiments that record negative results.
Compared with this lab: An example of publishing failures, which our benchmark also does.
Dhiman
The advantage of real topology disappears under shared initialisation and degree preserving nulls.
Compared with this lab: A warning our benchmark takes seriously: the rewired control keeps each neuron's number of connections.
Community demos from 2025 and 2026, many of them started after the MaleCNS release.
Ahmet Barış Günaydın
FlyWire brain, MANC nerve cord and flybody body in WebGPU; the player fires descending neurons to reach a target, with deterministic replays.
Compared with this lab: Technically the closest peer: brain and body in the browser. No control circuits.
vaibhavkedarisetti, syn-ack-ai
The whole FlyWire model in a Web Worker with 3D neurons, a raster, a circuit inspector, stimuli and lesions.
Compared with this lab: Whole brain in the browser without bodies or tasks.
Chris J. Zhang
The C. elegans connectome on WebGPU driving a worm on an agar plate, with touch and food.
Compared with this lab: The worm side of what this lab does, without controls.
Alex Wormuth (nftechie)
MaleCNS with dopamine gated plasticity plays Doom; reports honestly that one version failed its validation.
Compared with this lab: Adds learning inside the circuit, which this lab does not.
Mert Cobanov
An 80 neuron fly circuit plays the Chrome dinosaur game.
Compared with this lab: Inspired our fly runner; trained readout, no controls.
Seen through community lists only.
Community lists
Directories of about 130 fly connectome projects, updated in September 2026.
Compared with this lab: Where this lab should be listed.
The wiring diagrams everybody builds on.
FlyWire consortium, Princeton
The adult female fly brain, about 139,000 neurons, served on Codex.
Compared with this lab: The source of every fly circuit in this lab.
Janelia and Google
Brain and nerve cord of a male fly, more than 166,000 neurons; public release on 3 September 2026.
Compared with this lab: Next in our library, with the nerve cord for the body loop.
Harvard (htem) and partners
Female brain and nerve cord, about 160,000 neurons (Nature, June 2026).
Compared with this lab: A second route to a brain plus nerve cord model.
IARPA programme, Allen Institute and partners
One cubic millimetre of mouse visual cortex with functional recordings (Nature 2025).
Compared with this lab: A mammalian dataset on our planned list.
Groups working towards emulation, validation and open brain data.
Non-profit
BrainGenix NES simulator and the Brain Emulation Challenge with synthetic ground truth tissue; argues for validating internal dynamics, not just tasks.
Compared with this lab: Shares our controls first view; our NES bridge connects to their simulator.
Maximilian Schons and co-authors
The field's reference review of what emulation would take.
Compared with this lab: Background reading for the method page.
Non-profit research organisation
PRISM barcoding and expansion microscopy for cheaper whole mouse brain connectomes.
Compared with this lab: Could supply future datasets.
Successor of the Blue Brain Project
Open virtual labs for detailed biophysical brain simulation, launched March 2025.
Compared with this lab: Detailed neuron models where we use one simple model.
Allen Institute
Standardised cell data at scale, launched November 2025.
Compared with this lab: Cell type data that could refine our neuron model.
Biological computing: real tissue with unknown wiring, the mirror image of known wiring in a computer.
Cortical Labs, Melbourne
Cultured neurons played Pong (2022); the CL1 biocomputer followed in 2025 and about 200,000 human neurons played Doom in 2026.
Compared with this lab: Our Pong task invites a direct comparison between wetware and a mapped circuit.
FinalSpark, Switzerland
Remotely accessible human brain organoids for computing experiments.
Compared with this lab: Living tissue as a service; nothing about wiring is known.
Status and details change quickly in this field. If a project is missing or described wrongly, open an issue on GitHub.