ElectroPhysiomeGAN: Generation of Biophysical Neuron Model Parameters from Recorded Electrophysiological Responses

Area of Science:

• Neuroscience
• Computational Biology
• Biophysics

Background:

• Advances in connectomics and electrophysiology necessitate detailed neuron models.
• ElectroPhysiome models integrate network connectivity and cellular dynamics for simulating neuronal activity.
• The nematode C. elegans offers a tractable system for ElectroPhysiome studies due to its well-mapped connectome and available electrophysiological data.

Purpose of the Study:

• To develop a deep generative estimation method for accurately and rapidly inferring neuron model parameters from electrophysiological recordings.
• To enable the creation of detailed ElectroPhysiome models by estimating parameters for the Hodgkin-Huxley neuron model (HH-model).

Main Methods:

• Developed ElectroPhysiomeGAN (EP-GAN), a deep learning method combining Generative Adversarial Network (GAN) and Recurrent Neural Network (RNN) encoder.
• Trained EP-GAN to generate over 170 parameters for the HH-model using neuron membrane potential responses and steady-state current profiles.
• Validated EP-GAN on 200 simulated and 9 experimentally recorded neurons from C. elegans.

Main Results:

• EP-GAN efficiently generates extensive HH-model parameters (>170) for neurons with graded potential responses.
• Demonstrated high accuracy and inference speed compared to existing methods for parameter estimation.
• Showcased flexibility by allowing parameter inference from partial membrane potential and current data with arbitrary clamping protocols.

Conclusions:

• EP-GAN provides a powerful and efficient tool for estimating neuron model parameters, facilitating the construction of detailed ElectroPhysiome models.
• The method's speed and accuracy advance the simulation of neuronal network dynamics and cellular functions.
• EP-GAN's adaptability to partial data and varied protocols broadens its applicability in neuroscience research.