Reduced-Dimension, Biophysical Neuron Models Constructed From Observed Data

Area of Science:

• Computational neuroscience
• Applied mathematics
• Nonlinear dynamics

Background:

• Hodgkin-Huxley (HH) models offer detailed neuron simulations but are computationally intensive.
• Laboratory experiments provide real-world neuron data but require complex analysis.
• Existing reduced-dimension neuron models may not fully capture biophysical complexity.

Purpose of the Study:

• To develop a data-driven method for constructing accurate, reduced-dimension neuron models.
• To forecast neuron properties using only observable data, such as membrane voltage.
• To enable efficient construction and analysis of large neuronal networks.

Main Methods:

• Utilized nonlinear dynamics and interpolation techniques.
• Employed time-delay embedding to reconstruct dynamics from observable data (e.g., V(t)).
• Constructed discrete-time dynamical rules for forecasting.

Main Results:

• Data-driven forecasting (DDF) models accurately predict neuron properties beyond the training data.
• DDF models are reduced-dimension, focusing on observables like voltage, yet encode detailed biophysical information.
• Successfully predicted neuron behavior from both simulated HH models and experimental data.

Conclusions:

• DDF models offer a computationally efficient alternative to detailed HH models for network simulations.
• These models can be integrated into biophysically connected neuronal networks, replacing simpler models like integrate-and-fire.
• DDF facilitates the exploration of larger and more complex biological neural networks.