FastFly

GPU-accelerated simulator for the complete Drosophila melanogaster (fruit fly) brain connectome — 139,255 neurons and 54.5 million synapses — targeting real-time or faster performance on a single consumer NVIDIA GPU.

How it works

• Neuron model: Leaky Integrate-and-Fire (LIF), validated by arXiv:2404.17128
• Connectivity: CSR sparse format with FP16 synaptic weights
• Spike propagation: Push model — only processes synapses of neurons that actually fired
• Spike detection: Warp ballot intrinsics for bit-packed spike flags
• Load balancing: Warp-per-spike with grid-stride loop

Data sources

This simulator uses real connectome data from the FlyWire project — a collaborative effort to map every neuron and synapse in an adult Drosophila melanogaster brain from electron microscopy imagery.

Synaptic connectivity

• Data: Connectivity_783.parquet and Completeness_783.csv (FlyWire materialization v783)
• Repository: philshiu/Drosophila_brain_model
• Paper: Shiu PK, Sterne GR, Spiller N, et al. "A Drosophila computational brain model reveals sensorimotor processing." Nature 634, 210–219 (2024). doi:10.1038/s41586-024-07763-9
• Contents: 139,255 neurons, 54.5M synaptic connections with signed excitatory/inhibitory weights in CSR sparse format

Neuron annotations (cell types, positions, neurotransmitters)

• Data: Supplemental_file1_neuron_annotations.tsv
• Repository: flyconnectome/flywire_annotations
• Paper: Schlegel P, Yin Y, Bates AS, et al. "Whole-brain annotation and multi-connectome cell typing of Drosophila." Nature 634, 139–152 (2024). doi:10.1038/s41586-024-07686-5
• Contents: Cell type classifications (super_class, cell_class, cell_type), neurotransmitter identity, laterality, nerve assignments, and 3D soma positions in FAFB voxel coordinates (4×4×40 nm resolution)

Underlying electron microscopy volume

• Dataset: FAFB (Full Adult Fly Brain)
• Paper: Zheng Z, Lauritzen JS, Perlman E, et al. "A Complete Electron Microscopy Volume of the Brain of Adult Drosophila melanogaster." Cell 174(3), 730–743 (2018). doi:10.1016/j.cell.2018.06.019

FlyWire connectome

• Platform: flywire.ai · Codex browser
• Paper: Dorkenwald S, Matsliah A, Sterling AR, et al. "Neuronal wiring diagram of an adult brain." Nature 634, 124–138 (2024). doi:10.1038/s41586-024-07558-y

Neuron model validation

• Zhang X, Yang P, Feng J, et al. "Network Structure Governs Drosophila Brain Functionality." arXiv:2404.17128 (2024). Demonstrates that network structure dominates over neuron model choice, validating the use of LIF for whole-brain simulation.

Requirements

• CUDA path (C++): NVIDIA CUDA Toolkit 12.x
• Python path: Python 3.10+, CuPy (pip install cupy-cuda12x)
• Web visualizer: FastAPI, uvicorn (pip install fastapi uvicorn[standard])

Quick start

1. Download the connectome

pip install pandas pyarrow numpy requests
python download_connectome.py

This downloads the FlyWire v783 data and produces flywire_v783.bin.

2a. Run the CUDA simulator (C++)

build.bat                               # compile (requires nvcc)
flywire_sim.exe --data flywire_v783.bin  # run with real connectome
flywire_sim.exe                          # or run with synthetic data

2b. Run the Python/CuPy simulator

pip install cupy-cuda12x
python flywire_sim.py --data flywire_v783.bin
python flywire_sim.py                    # synthetic data fallback

3. Web visualizer

pip install fastapi uvicorn[standard]
python app_server.py --data flywire_v783.bin
# Open http://127.0.0.1:8000

Project structure

File	Description
flywire_sim.cu	Standalone CUDA C++ simulator
flywire_sim.py	Python/CuPy simulator (runtime-compiled CUDA kernels, no Visual Studio needed)
sim_engine.py	Simulation engine used by the web server
app_server.py	FastAPI web server with WebSocket-based live visualizer
download_connectome.py	Downloads FlyWire v783 data and converts to binary format
download_metadata.py	Downloads neuron annotation metadata
build.bat	Build script for the C++ simulator (targets RTX 3080 Ti / SM 8.6)

License

MIT