Drosophila brain model
Research code for the Shiu et al. connectome-based leaky integrate-and-fire model, including activation/silencing experiments, notebooks, and FlyWire data configuration.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.Model predictions are tested against experiments in real flies — optogenetic activation, calcium imaging, and feeding and grooming behaviour — rather than against anatomy alone.
https://www.biorxiv.org/content/10.1101/2023.05.02.539144v1Every LIF constant is stated with a literature citation, and the membrane/synapse equations are written out explicitly.
https://github.com/philshiu/Drosophila_brain_model/blob/91bdd1e7dcf193f3e7ca5a8933497fcef63b7960/model.py#L15-L53Synaptic weights are assigned once from the FlyWire connectivity table times the free parameter w_syn, and no rule updates them thereafter.
https://github.com/philshiu/Drosophila_brain_model/blob/91bdd1e7dcf193f3e7ca5a8933497fcef63b7960/model.py#L175-L183The only code that writes to syn.w sets it to zero to model optogenetic silencing of selected neurons - an intervention, not a plasticity rule.
https://github.com/philshiu/Drosophila_brain_model/blob/91bdd1e7dcf193f3e7ca5a8933497fcef63b7960/model.py#L108-L127The accompanying preprint reports that model predictions were tested against real flies using optogenetic activation, calcium imaging and behavioural experiments on feeding and grooming circuits.
https://www.biorxiv.org/content/10.1101/2023.05.02.539144v1
This is the reference implementation of the Shiu et al. whole-brain model and the strongest neuron model in the batch: a Brian2 LIF network over the FlyWire connectome with every constant sourced to a published measurement (v_rest/v_reset -52 mV, threshold -45 mV, 20 ms membrane constant from Kakaria & de Bivort 2017, 5 ms alpha synapse from Juergensen et al., 2.2 ms refractory from Lazar et al., 1.8 ms delay from Paul et al.), and one free parameter w_syn = 0.275 mV per synapse. Not L2: syn.w is written in exactly two places in the whole repository. Once at model.py L183, assigned from the connectivity table; and once inside silence(), which zeros the outgoing weights of chosen neurons to model optogenetic silencing. That second write is an experimenter's manipulation, not plasticity, and it is the one thing here easy to mistake for a learning rule when grepping. There is no dopamine term, no eligibility trace and no learning rate anywhere in model.py or utils.py. Validated: the accompanying preprint does not stop at anatomy. Model predictions about which feeding-region neurons drive motor-neuron firing were tested in live flies by optogenetic activation and behavioural assay, with calcium imaging on the shared sugar/water appetitive pathway and behavioural tests on the antennal grooming circuit. On this site the level and the mark are deliberately separate questions, so a model with no plasticity stays at L1 and still carries the mark - this project is the clearest illustration of why the two were split apart.
Record
Nearby
All projects →FlyWire whole-brain LIF implementation based on Shiu et al., with multiple simulation backends and benchmarking tools. This repository supplies the neural model; it is not a complete embodied demo package.
Anatomically detailed MuJoCo fruit-fly body, walking and flight environments, and reinforcement-learning examples. A body and control platform.
Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.