Drosophila Whole-Brain Spiking Neural Network Simulation

A computational reproduction of the whole-brain Drosophila melanogaster spiking neural network model based on the FlyWire connectome.

Overview

This repository reproduces the computational model from the Nature paper:

A Drosophila computational brain model reveals sensorimotor processing Philip K. Shiu, Gabriella R. Sterne, Nico Spiller, Romain Franconville, et al. Nature 634, 210-219 (2024) DOI: 10.1038/s41586-024-07763-9

This implementation is modified from the original repository by Philip Shiu and collaborators.

Model Description

The model implements a leaky integrate-and-fire (LIF) spiking neural network of the entire adult Drosophila central brain, containing:

• 127,400 neurons derived from the FlyWire connectome
• ~50 million synaptic connections with excitatory/inhibitory weights
• Biologically realistic parameters:
  • Membrane time constant: 20 ms
  • Synaptic time constant: 5 ms
  • Refractory period: 2.2 ms
  • Resting potential: -52 mV
  • Threshold potential: -45 mV
  • Synaptic delay: 1.8 ms

Neuron Dynamics

The dynamics follow the LIF equations:

dv/dt = (v_rest - v + g) / tau_m
dg/dt = -g / tau_syn

Where v is membrane potential, g is synaptic conductance, tau_m is membrane time constant, and tau_syn is synaptic time constant.

Features

• Whole-brain simulation: Simulates all 127,400 neurons with their complete connectivity
• Sensorimotor processing: Models gustatory (taste) and mechanosensory pathways
• Multiple input modalities: Sugar, bitter, water sensing, and Johnston's organ neurons (JON)
• Silencing experiments: Supports inhibiting specific neuron populations
• GPU acceleration: Built on JAX for efficient parallel computation

Installation

Prerequisites

• Python 3.9+
• CUDA-compatible GPU (recommended for full simulations)

Dependencies

pip install BrainX pandas seaborn

Usage

Basic Simulation

from drosophila_whole_brain import Network, run_one_exp
import brainstate
import brainunit as u

# Define neurons to stimulate (sugar-sensing gustatory receptor neurons)
neu_sugar = [
    720575940624963786, 720575940630233916, 720575940637568838,
    # ... additional FlyWire neuron IDs
]

# Run simulation
firing_rates = run_one_exp(
    neurons_to_excite=[{'ids': neu_sugar, 'rate': 150 * u.Hz}],
    duration=1000 * u.ms,
    dt=0.1 * u.ms,
)

Running with the Network Class

import brainstate
import brainunit as u
import numpy as np
from drosophila_whole_brain import Network

with brainstate.environ.context(dt=0.1 * u.ms):
    net = Network(
        path_neu='./2023_03_23_completeness_630_final.csv',
        path_syn='./2023_03_23_connectivity_630_final.parquet',
        neuron_to_excite=[
            {'ids': neu_sugar, 'rate': 150 * u.Hz},
        ],
    )
    brainstate.nn.init_all_states(net)

    # Run for 1000 timesteps
    indices = np.arange(10000)
    spks = brainstate.transform.for_loop(
        lambda i: net.step_run(i, ret_val='spike'),
        indices
    )

Repository Structure

.
├── README.md                                    # This file
├── drosophila_whole_brain.py                    # Core simulation module
├── brainstate_figures.ipynb                     # Jupyter notebook reproducing paper figures
├── 2023_03_23_completeness_630_final.csv        # Neuron metadata (FlyWire IDs)
├── 2023_03_23_connectivity_630_final.parquet    # Synaptic connectivity matrix
└── sez_neurons.pickle                           # SEZ neuron type classifications

Reproducing Paper Figures

The brainstate_figures.ipynb notebook reproduces key figures from the paper:

• Figure 1D: Whole-brain activity when activating sugar GRN
• Figure 1E: Which neurons activate MN9 motor neuron
• Figure 1F: Silencing experiments
• Figure 2: Which cell types activate MN9
• Figure 3: Sugar + bitter / sugar + IR94e interactions
• Figure 4: Water GRN activation and interactions
• Figure 5: Johnston's organ neuron (JON) experiments

Run the notebook:

jupyter notebook brainstate_figures.ipynb

Data Sources

The connectome data is derived from:

• FlyWire - A complete whole-brain connectome of Drosophila
• Data version: 2023_03_23, completeness threshold 630

Citation

If you use this repository, please cite:

This Repository

@software{wang2025drosophila,
  author       = {Wang, Chaoming},
  title        = {chaobrain/drosophila\_whole\_brain\_snn\_simulation: Release Version 0.1},
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.17791458},
  url          = {https://doi.org/10.5281/zenodo.17791458}
}

Original Paper

@article{shiu2024drosophila,
  title     = {A Drosophila computational brain model reveals sensorimotor processing},
  author    = {Shiu, Philip K. and Sterne, Gabriella R. and Spiller, Nico and
               Franconville, Romain and Sandoval, Andrea and Zhou, Joie and
               Bhardwaj, Neha and Prieto-Godino, Lucia L. and Bhalla, Upinder S. and
               others},
  journal   = {Nature},
  volume    = {634},
  number    = {8032},
  pages     = {210--219},
  year      = {2024},
  publisher = {Nature Publishing Group},
  doi       = {10.1038/s41586-024-07763-9}
}

Original Implementation

@misc{shiu2024drosophila_code,
  author    = {Shiu, Philip K.},
  title     = {Drosophila brain model},
  year      = {2024},
  publisher = {GitHub},
  url       = {https://github.com/philshiu/Drosophila_brain_model}
}

License

This project is licensed under the Apache License 2.0 - see the source files for details.

Acknowledgments

• FlyWire consortium for the connectome data
• Philip Shiu and collaborators for the original model implementation
• BDP Ecosystem for the brainstate/brainunit/brainevent frameworks