Fruit Fly Simulator

How to play

1. The left room, with the banana, is the rearing room. Tap a fly and it is carried through the tube in the middle to the experiment room on the right, and its real brain switches on. The magnifier at the top right shows it up close.
2. Tap anything in the row below and it appears in the experiment room. A filled button means it is in the room now. Ball rig, Yeast, Spider and Courtship stay until you tap them again (you can put in several at once); Sugar, Bitter sugar, Shadow, Brush, Sound and Vinegar are used once on the fly you brought and then go away.
3. The Ball rig tethers the chosen fly on a ball (when it walks the ball turns, and a virtual path is drawn at the top left). Yeast makes females that eat it lay far more eggs. The Spider creeps up on a fly on the floor and pounces. With Courtship, a male brought to a dish follows a female and sings by vibrating a wing. Sugar and Bitter sugar go as a drop in front of the fly on the floor, or by glass capillary to the tip of the mouthparts on the rig. Shadow is a hand shadow on the floor, or a black disc on a side screen on the rig (left and right in turn). The Brush tickles the head bristles, Sound brings out a speaker that clicks, and Vinegar sprays odour. Vinegar makes this model's brain run away.
4. Light on, above the brain, switches on only command neurons that go to the body, with red light. You can make the fly walk backward or forward, turn or jump without any sense input. Tap again to switch it off.
5. For every stimulus, and every spider pounce, it records which body signal crossed its threshold first and how many ms it took. See the table and a summary per stimulus under Results above the brain.
6. Tap the brain and the neuron type at that spot switches on for a moment; press and hold to keep it on. Drag the brain to turn it. Zoom into the case toward a spot with the wheel or two fingers, and drag to move around while zoomed.
7. In the case, one day is 6 seconds. Females lay eggs on the banana; full-grown larvae crawl out of it, climb the front glass and pupate, and new flies emerge there. The banana shrinks as it is eaten and rots. When it runs out, swap it with New banana at the top right.
8. Tap Release and the fly goes back to the rearing room and its brain switches off.

The responses on screen are model computations. They do not predict real fly behaviour.

How this brain runs

• The brain is FlyWire's wiring diagram of an adult female fruit-fly brain, reconstructed from electron microscopy (version 783). Each of its 138,639 neurons is drawn as dots at its real cell body and at one point on its branches; the tail of each spark is the line joining the two.
• Only the chosen fly moves by this brain. The rest rest, walk, fly and gather at food by simple rules, because running dozens of brains at once on a phone is too much. The flight to the experiment room and back is done by the simulator. Once the fly lands on the room's floor, its brain's walking, turning and jump signals move its body (forward 16 mm/s, backward 9 mm/s, turning 3.2 rad/s, a jump when the giant fibre passes 20 Hz · all set values). If it jumps onto the glass, the simulator puts it back on the floor after a moment.
• The ball rig imitates the tethered-on-a-ball setup used in fly labs. A fly glued by the thorax to a fine pin walks on an air-supported ball; the ball turns, and the path walked is worked back from its rotation (the virtual path). Here the forward-walking neurons (DNp09) at 100 Hz walk it at 16 mm/s, moonwalker (MDN) at 100 Hz backs it up at 9 mm/s, and a 90 Hz left-right difference in the turning neurons (DNa01·02) turns it at 3.2 rad/s. When the giant fibre passes 20 Hz it spreads its wings to jump but, tethered, cannot leave. All are values this simulator sets.
• The jumping spider picks a fly on the room's floor, turns to face it, creeps closer in stops and starts, crouches 18–42 mm away and leaps. The leap takes the distance divided by 0.45 m/s (about 40–95 ms). While it creeps, and only while it moves, the looming neurons of the eye facing it get a weak input in step with how fast its image grows (stronger and on more neurons as it gets closer; nothing while it holds still). Jumping spiders move in stops and starts precisely to be noticed less. Flies rarely flee while it creeps (2 times in 40, within 22 mm). During the leap, the looming neurons of the eye facing the spider get an input rising with the square of the leap's progress (up to 150 Hz). The fly escapes if its giant fibre passes 20 Hz 5 ms before the spider's legs reach it (at 85% of the leap; both set values). Run 40 times per leap time, the same computation escaped 45 ms 0/40, 55 ms 6/40, 65 ms 21/40, 75 ms 30/40, 85 ms 38/40 and 95 ms 39/40. The other flies escape at these rates.
• The spoonful of yeast is a rule. On each flight, females fly to the yeast and eat with 17% chance, males 5%. Females that ate yeast go back to the banana to lay, with up to three times the chance of laying and more eggs even on a crowded banana (real females also need protein to make eggs). The numbers are set values. Run twice for 3 minutes, the case with yeast laid nearly twice as many eggs, ran out of banana sooner and collapsed first.
• Courtship is a rule too. The male turns toward the female, follows her and spreads and vibrates the wing on her side (in real flies this vibration is the song). After about 5–11 seconds of song the female accepts with 70% chance and they mate (the real 20 minutes or so shortened to 12–18 seconds); otherwise she runs off. In this brain model, switching on the antennal hearing neurons makes the giant fibre trigger a jump, so a female run on the brain would hear the song and leap away. Real females don't, so she moves by rules instead of the brain.
• The tools switch on real sensory neurons. Sugar: the sugar-taste neurons at the tip of the mouthparts (LB3c, 32); bitter sugar adds the bitter-taste neurons (42); shadow: the looming neurons of that eye (LPLC2·LC4); brush: the head-bristle neurons (305); sound: the hearing neurons of Johnston's organ in both antennae (387); vinegar: odour neurons (DM1). Each switched-on neuron fires at random 150 times a second (head bristles 80), so the same stimulus differs a little every time.
• The results table watches seven signals to the body (forward, backward, left turn, right turn, jump, proboscis, grooming), their firing rates averaged over 80 ms and read every 0.5 ms, and records as the first response the signal that first crosses its threshold within (stimulus length + 0.6 s) of the stimulus. Thresholds are 20 Hz for jump and proboscis and 30 Hz for the rest (set values).
• The same computation as this page, run 10 times per stimulus: no stimulus → 0/10. Sugar → proboscis extension 10/10 (mean 61 ms). Bitter sugar → no response 10/10. Shadow → jump 10/10 (mean 19 ms), then a turn away 10/10. Brush → grooming 10/10 (mean 23 ms). Sound → jump 10/10 (mean 75 ms). Vinegar → runaway 10/10 (mean 50 ms). Command neurons switched on by light → that movement 10/10 (mean 17–25 ms).
• Light on imitates an optogenetics experiment: it assumes a light-sensitive protein in just one type of descending neuron and switches on only those. Moonwalker neurons (MDN) are known to drive walking backward, DNp09 walking forward, DNa01·02 turning to their side and the giant fibre jumping. The neurons switched on are the very ones read for the body's response, so the movement comes at once. They send their signals mostly to the ventral nerve cord (VNC, not in this wiring diagram), so the sparks inside the brain are small.
• Wind-sensing and water-taste neurons were tried too. A few kinds of descending neurons do switch on, but what those neurons do in the body is not well known yet, so they are not turned into movement.
• The neuron model is the simple spiking model of Shiu et al. (2024, Nature), rewritten for this page. Run on the same sugar experiment as the original (Brian 2), firing rates per neuron are nearly the same (correlation 0.998). To run on phones it uses only the 2.7 million connections of 5 or more synapses (63% of all synapses) and steps every 0.5 ms.
• There is no body model. The rules that turn the firing of a few descending neuron types into ball turning, jump attempts, proboscis extension and grooming are set by this simulator. Counts in the results table are counts from this model, not the chance that a real fly would do so, and they do not predict real fly behaviour.
• Switching on odour or temperature neurons sets the whole brain ablaze without stopping, in the original model too. Real flies don't do this. Here it is shown as a 'runaway', and after 3 seconds that fly's brain is reset.
• The life cycle follows real development times at 25 °C: egg 1 day, larva 4 days (moulting three times and finally crawling out of the food), pupa 4 days, fast-forwarded at 6 seconds a day. Adults live 30–70 days, and lay far fewer eggs than real flies (dozens a day) so the case doesn't overflow. Crowded larvae die in numbers, as in real life. This life cycle and the flock's behaviour are rules.
• Each day the banana shrinks 0.8% from rotting, 0.05% per feeding larva and 0.02% per adult. Below a quarter, fewer eggs are laid. When it runs out, young larvae starve within a day, while late third-instar larvae pupate on what they have eaten (critical weight). Adults last 2–5 days (females longer); real fruit flies with only water also live about 2–3 days.
• Choose a male (dark tip to the abdomen, a little smaller) and it still runs on the same female wiring diagram, because the published whole-brain wiring diagram is a female's.
• The case is two cubes of about 21 cm joined together (inside 42 × 21 × 20 cm), with a glass tube 3.6 cm across and 4.6 cm long set in the middle of the glass wall between them. Left is the rearing room with the banana; right is the experiment room, where only what you choose goes in. The banana sits against the front glass, so most full-grown larvae climb the nearest front glass to pupate. Flock flies sometimes (3.5% of flights) visit the experiment room, and from there return with 34% chance on each flight. Walking flies and larvae can't cross the glass wall, so eggs, larvae and pupae appear only in the left room. Flies, eggs, larvae and pupae are drawn about twice real size to be seen from afar, and the rig's ball and the spider are scaled up as much. Colours follow wild-type Drosophila melanogaster: a tan body, brick-red eyes, black bands across the abdomen, pale when newly emerged and darkening over a day or so.

Sources: Dorkenwald et al. 2024 · Schlegel et al. 2024 (Nature), FlyWire (CC BY 4.0), flywire_annotations · Shiu et al. 2024 (Nature), Drosophila_brain_model (MIT)