What does the feeding motor neuron depend on?
Watching which neurons fire alongside a motor neuron is not the same as knowing which ones it needs. This page runs both tests on the whole fly brain model, then checks the model's predictions against real-fly behavior. No neural recordings are compared yet.
Correlation is not causation
Each dot is one neuron that fires when the model fly tastes sugar. Across: how closely its firing tracks the rostrum motor neuron MN9 over time. Up and down: how much MN9 drops when that one neuron is silenced in the full-brain model. Click a dot to open it in the lab.
Both axes are model activity, not recordings. Correlation: Pearson r of 2 ms binned spike counts, averaged over 10 trials, at the best lag between −20 and +20 ms, minus a trial-shuffled predictor so that firing merely caused by the shared sugar input does not count. Dots inside the grey band are within the 95th percentile of a time-shifted null. Real calcium imaging cannot resolve lags this short. Silencing: mean MN9 rate over 10 trials with the neuron's outgoing synapses set to zero, as in Shiu et al. 2024.
Largest drop when silenced alone
Fires most in step with MN9 (corrected)
Silence a whole cell type
Silencing one neuron at a time can hide redundancy: most cell types come in left–right pairs. Here each cell type is silenced on both sides at once, and compared with silencing the same number of randomly chosen sugar-active neurons.
Harder sugar-neuron drive, more MN9
MN9's firing rate as the 20 sugar neurons are driven at higher Poisson rates. This is neuron drive, not sugar concentration. The shape is a threshold, then a climb.
Bitter vetoes sugar
Bitter neurons on the same side as the sugar neurons (cell_sub_class “bitter”). In real flies bitter taste suppresses feeding; the model reproduces the veto, given predicted transmitter signs and equal 150 Hz drive to 32 bitter and 20 sugar neurons. Tested at one ratio only.
Does the specific wiring matter?
Scorecard against real flies
Shiu et al. switched on 106 taste-circuit cell types one at a time in living flies with light (optogenetics) and recorded whether the fly extended its proboscis. Here the same 106 activations run through this lab's engine, and each prediction is scored against what the real flies did (Fig. 2, Supplementary Table 3).
Second scorecard: which cell types does feeding need?
Shiu et al. also silenced ten cell types in living flies (light-gated GtACR1) and asked whether sugar still triggered feeding. The same test runs here with the paper's rule: a type is “required” if silencing it cuts MN9 to 80% of control or less at any sugar drive from 50 to 120 Hz.