Estimating latent neuronal nonlinear dynamics by sequential Monte Carlo method and sparse modeling

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Biophysics

Background:

• Accurate neuronal models are crucial for understanding brain function.
• Estimating these models from time-series data is challenging due to partial observability.

Purpose of the Study:

• To develop a data-driven sparse modeling method for estimating neuronal latent variables and electrical properties.
• To extract essential membrane currents from partially observable time-series data.

Main Methods:

• Derived a nonlinear state-space model from a conductance-based neuron model.
• Developed a sparse modeling-based expectation-maximization (EM) algorithm.
• Utilized sequential Monte Carlo (SMC) for approximating posterior distributions under noisy observations.

Main Results:

• The proposed method successfully estimates latent variables and biophysical parameters.
• Essential membrane currents were effectively extracted, distinguishing necessary from unnecessary ones.
• Simulations demonstrated accurate extraction of latent nonlinear neuronal dynamics.

Conclusions:

• The data-driven sparse modeling approach offers a robust method for neuronal model estimation.
• This technique enhances the analysis of complex neuronal behavior from limited observational data.
• It provides a pathway for more accurate characterization of neuronal electrical properties.