Efficient models of cortical activity via local dynamic equilibria and coarse-grained interactions

Area of Science:

• Computational neuroscience
• Systems neuroscience
• Biophysics

Background:

• Biologically detailed neural network models are computationally intensive due to complex neuronal interactions and unknown parameters.
• Simplified models sacrifice biological realism, limiting their evaluative power.
• A multiscale approach offers a balance between biological accuracy and computational tractability.

Purpose of the Study:

• To present a novel multiscale, coarse-grained (CG) model for simulating brain circuitry.
• To achieve a balance between biological realism and computational efficiency in neural modeling.
• To reduce the computational cost of large-scale neural network simulations.

Main Methods:

• Developed a coarse-grained model representing neuronal groups as 'pixels'.
• Alternately updated dynamics at intra- and inter-pixel scales until convergence.
• Modeled intrapixel dynamics as a single system driven by external inputs, leveraging cortical anatomical similarities.
• Precomputed and tabulated local responses to accelerate simulations.

Main Results:

• The model reproduces key features of large-scale network models, such as neuronal orientation selectivity.
• Achieved significant computational cost reduction compared to direct multiscale simulations.
• Demonstrated the methodology using a model of the primate visual cortex.

Conclusions:

• The proposed multiscale, coarse-grained modeling approach offers a computationally efficient alternative for simulating brain circuitry.
• This method preserves essential local biological details while enabling large-scale network analysis.
• The model effectively captures functional properties like orientation selectivity with reduced computational overhead.