Parameter Optimization Using Covariance Matrix Adaptation-Evolutionary Strategy (CMA-ES), an Approach to Investigate...

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Biophysics

Background:

• Developing accurate biophysically realistic single neuron models is challenging.
• Computational models are crucial for predicting how neuronal mechanisms affect network activity.

Purpose of the Study:

• To present user-friendly software for automatic parameter optimization of computational neuronal models.
• To ensure compatibility with common neural simulation software, specifically the MOOSE simulator.

Main Methods:

• The software utilizes a declarative format for MOOSE models and accepts experimental data in specified file formats.
• A customizable fitness function, a weighted combination of feature differences, is employed.
• The covariance matrix adaptation-evolutionary strategy is used for robust optimization in complex fitness landscapes.

Main Results:

• The software successfully generated models for four neuron types (spiny projection and globus pallidus subtypes) by fitting to current clamp data.
• Optimization converged within 1,600-4,000 model evaluations.
• Analysis of optimized parameters revealed distinct differences between neuron subtypes, aligning with existing experimental findings.

Conclusions:

• The developed software offers an accessible, automated method for determining neuron channel parameters.
• This approach can be applied to various neuron subtypes using experimental recordings to elucidate ionic differences.