Data Assimilation Methods for Neuronal State and Parameter Estimation.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Mathematical Biology

Background:

• Data assimilation (DA) techniques are established in climate and weather science but are emerging in neuroscience.
• DA methods offer powerful tools for estimating unobserved variables and parameters in complex biological models.
• Neuronal models, like the Morris-Lececar model, exhibit diverse behaviors based on parameter values and bifurcation structures.

Purpose of the Study:

• To demonstrate the application of data assimilation algorithms for parameter and state estimation in neuronal models.
• To provide accessible code implementations of key DA techniques for neuroscience research.
• To evaluate the ability of DA methods to infer correct neuronal excitability regimes from voltage data.

Main Methods:

• Implementation of sequential (Unscented Kalman Filter) and variational (4D-Var) data assimilation algorithms.
• Utilizing computer code to apply DA methods to the Morris-Lecar neuronal model.
• Inferring model parameters and assessing estimation success using voltage traces from different excitability regimes.

Main Results:

• DA algorithms successfully estimated parameters of the Morris-Lecar model from single voltage traces.
• The methods accurately identified the correct bifurcation structure and excitability regime, even from poor initial parameter estimates.
• The geometric structure of inferred models was shown to be a useful qualitative measure of estimation success.

Conclusions:

• Data assimilation techniques are effective for nonlinear state and parameter estimation in neuroscience.
• DA offers a robust approach to characterizing neuronal dynamics and model parameters.
• Further extensions of DA algorithms hold promise for advancing computational neuroscience research.