Jaxley

Jaxley enables large-scale gradient-based optimization of detailed biophysical neuron models through differentiable simulation and GPU acceleration.

Key Features:

• Differentiable simulation: Jaxley uses automatic differentiation to compute gradients for optimizing biophysical model parameters.
• GPU acceleration: Jaxley leverages GPU computing to accelerate simulations and training of large-scale neuron models.
• Multi-compartment neuron support: Jaxley simulates detailed biophysical models including morphologically realistic neurons with 100,000 parameters.
• Network simulation: Jaxley supports simulation and training of recurrent neural networks composed of biophysical neurons.
• Multi-objective fitting: Jaxley can optimize models to match voltage recordings or two-photon calcium imaging data.

Scientific Applications:

• Electrophysiological data fitting: Jaxley learns biophysical neuron parameters to match intracellular voltage recordings.
• Calcium imaging data fitting: Jaxley optimizes model parameters to match two-photon calcium imaging recordings.
• Working memory modeling: Jaxley trains recurrent networks of biophysical neurons to perform working memory tasks.
• Computer vision with biophysical networks: Jaxley trains morphologically detailed neuron networks to solve computer vision tasks.
• Mechanistic investigation of neural computation: Jaxley builds data- and task-constrained models to study cellular mechanisms underlying neural computations across scales.

Methodology:

Jaxley implements differentiable simulation of biophysical neuron models using JAX, applying automatic differentiation and gradient descent for parameter optimization.