eonsystemspbc/fly-brain

What it solves

This project provides a computational emulation of the adult fruit fly (Drosophila) brain. It allows researchers to simulate how neural activity propagates through the fly's entire brain connectome, enabling the testing of hypotheses about sensorimotor processing by activating or silencing specific neurons and observing the resulting spike patterns.

How it works

The system implements a whole-brain leaky integrate-and-fire (LIF) model based on the FlyWire connectome, which consists of approximately 138k neurons and 5 million synapses. It supports two primary manipulations:

• Activation: Simulates optogenetic activation by triggering Poisson spiking in target neurons at a fixed frequency.
• Silencing: Simulates optogenetic silencing by setting all synaptic connections to and from target neurons to zero.

To ensure performance and accuracy, the project provides multiple simulation backends, including PyTorch, GeNN, NEST GPU, and Brian2, allowing users to run simulations on both CPUs and GPUs.

Who it’s for

It is designed for computational neuroscientists and researchers studying the Drosophila brain who want to simulate neural activity and validate their findings against experimental data.

Highlights

• Multi-Backend Support: Compatible with Brian2, Brian2CUDA, PyTorch, NEST GPU, and GeNN.
• Connectome-Scale: Based on the FlyWire v783 connectome (~138k neurons, ~5M synapses).
• Ground Truth Validation: Uses Brian2 (CPU) as a gold standard to measure the accuracy of GPU-accelerated backends.
• Flexible Manipulations: Supports both targeted activation and silencing of arbitrary neuron sets.