About WiredMind

What’s real and what’s simplified

Real

• The neurons. Each lesson uses 3,000 to 4,500 reconstructed neurons from the MaleCNS v1.0 (opens in a new tab) connectome, the whole central nervous system of one male fly, with the cell types its authors assigned.
• The synapse counts. The weight of every connection is the number of synapses the dataset reports between the two cells, at detection confidence 0.5 or above.
• The signs. A neuron’s predicted transmitter sets whether it excites or inhibits: acetylcholine excites, GABA and glutamate inhibit. Cells predicted to release dopamine, serotonin, octopamine, or an unclear transmitter have no effect in the model.
• The shapes and positions. Each neuron is drawn from its own skeleton in the dataset, where it really sits in the fly.

Simplified

• Small cuts. Each lesson runs a subcircuit of 3,000 to 4,500 neurons, not the whole nervous system. The smell circuit is four cell populations: receptors, projection neurons, Kenyon cells, and output neurons. The compass and escape circuits grow two hops out from their seed cell types and keep the most strongly connected partners. Connections of fewer than five synapses are dropped. Everything outside the cut is gone, including inhibition that would normally balance the network.
• One neuron model for every cell. Leaky integrate-and-fire point neurons with the same parameters everywhere, the published values of Shiu et al. (2024) (opens in a new tab): rest and reset −52 mV, threshold −45 mV, membrane time constant 20 ms, synaptic time constant 5 ms, 2.2 ms refractory period, 1.8 ms delay, 0.275 mV per synapse. No dendritic computation, no graded or non-spiking neurons, no neuromodulation, and no learning: synapses never change.
• No gap junctions. The connectome records chemical synapses only, so electrical synapses are absent. This matters most in the escape lesson: in a real fly the giant fiber’s link to the jump motor neuron is a mixed synapse that leans heavily on gap junctions. Here that link is its 90 chemical synapses.
• Nothing learns. The smell lesson shows where a fly stores odor memories, the synapses from Kenyon cells to output neurons, but the model has no dopamine neurons and no plasticity. The cut also leaves out the APL neuron that keeps Kenyon cell activity sparse.
• Idealized stimulation. Stimulate drives every neuron in a group with independent Poisson input (40 Hz for 200 ms in the smell and escape lessons, 60 Hz for 800 ms for the turn neurons), each event strong enough to fire the cell, much like optogenetic activation. A real odor activates a few receptor types in a pattern; the lesson’s smell fires all 1,861 receptor neurons in the cut. The swatter fires all 304 looming cells at once, where a real looming object recruits them as it grows.
• Silence is a hard clamp. A silenced group sits at rest, never spikes, and sends nothing. Real silencing tools such as Kir2.1 or tetanus toxin are partial, slower, and can leave electrical synapses working.
• Compass landmarks. The ring neurons receive no visual input in the model. Silencing them removes their inhibition of the compass, which is what lets the bump drift here. In a real fly they bring landmarks and other cues in from the senses.
• Timing. Latencies come from one uniform 1.8 ms delay and the membrane and synaptic time constants, not from recordings. They land in a plausible range, but nothing was tuned to match measured escape latencies. The viewer plays it all in slow motion, 0.2 to 2 ms of fly time per frame depending on the device, and stops simulating 150 ms after the last stimulus, because some cuts have excitatory loops that would otherwise echo forever without the inhibition left outside.
• Drawings, not morphology. Each neuron is drawn as a 12-point line along its skeleton, not its real branches. Flashes mark spikes. The moving dots are illustration, not conduction speed or the real path of a spike.

What a research tool would do differently

• Simulate the whole central nervous system, or cut it on principled boundaries with modeled inputs, and keep weak connections.
• Fit parameters by cell type to recordings, allow graded and non-spiking neurons, and use conductance-based or multi-compartment models where dendrites matter.
• Add gap junctions from other data, neuromodulators such as dopamine and octopamine, and plasticity, for example dopamine-gated depression at Kenyon cell to output neuron synapses.
• Drive sensory neurons the way the world does: odor-specific receptor patterns, and looming stimuli that recruit cells as the object grows.
• Treat the transmitter predictions and weights as uncertain, sweep them, run many seeds, and compare the results with electrophysiology and behavior.

Go to the source

• neuPrint (opens in a new tab): query MaleCNS v1.0 cell by cell and synapse by synapse.
• The MaleCNS paper (opens in a new tab): Berg, S., Beckett, I. R., Costa, M., Schlegel, P., Januszewski, M., et al. (2026). Sexual dimorphism in the complete Drosophila male central nervous system connectome. Cell 189, 5504–5526.e15.
• Shiu et al. (2024) (opens in a new tab): the whole-brain leaky integrate-and-fire model whose parameters WiredMind borrows.
• The source code (opens in a new tab): the data pipeline, the simulator, the lessons, and tests that check each lesson’s claims on the real circuits.

Where the neurons come from

Every neuron and synapse count comes from the MaleCNS v1.0 connectome (opens in a new tab) of the male fruit fly’s central nervous system, made by HHMI Janelia FlyEM (opens in a new tab), the University of Cambridge (opens in a new tab), the MRC Laboratory of Molecular Biology (opens in a new tab), and Google Research (opens in a new tab).

Berg, S., Beckett, I. R., Costa, M., Schlegel, P., Januszewski, M., et al. (2026). Sexual dimorphism in the complete Drosophila male central nervous system connectome. Cell 189, 5504–5526.e15. doi:10.1016/j.cell.2026.08.015 (opens in a new tab)

The data is licensed under CC BY 4.0 (opens in a new tab). WiredMind changes it: it cuts out three small circuits, drops connections of fewer than five synapses, and draws each neuron as a simplified line. The simulator uses the neuron model of Shiu et al. (2024) (opens in a new tab).

Who made this

WiredMind was built by Guy Dahan, a full-stack developer and SEO strategist in Tel Aviv, over one week in September 2026.

GitHub (opens in a new tab)·guy-dev.com (opens in a new tab)

Support

WiredMind is free and always will be. If it helped your class, coffee keeps the server humming.

☕ Buy me a coffee (opens in a new tab)

Privacy

No accounts. The only cookie remembers a language you pick from the language menu. Page views are counted with Vercel Web Analytics, which does not use cookies.