Nonlinear model predictive control of a conductance-based neuron model via data-driven forecasting.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Precise neural system control is crucial for brain-behavior research and therapeutic interventions.
• Model predictive control (MPC) offers a promising framework for managing complex neural dynamics, noise, and incomplete state information.
• Challenges include model selection, parameter constraints, and system synchronization in neural control.

Purpose of the Study:

• To demonstrate a data-driven approach for creating a nonlinear model of a neuron with limited observable data.
• To apply this model within a nonlinear MPC framework for precise control of neuronal activity.
• To address the challenge of controlling conductance-based models with unobservable states and parameters.

Main Methods:

• Utilized advanced data-driven forecasting techniques to build a nonlinear machine-learning model.
• Modeled a Hodgkin-Huxley type neuron using only observable membrane voltage.
• Assumed an unknown number of intrinsic currents within the neuron model.

Main Results:

• The developed model successfully learned the dynamics of different neuron types.
• The nonlinear MPC approach, using the learned model, could drive neurons to exhibit specific, researcher-defined spiking behaviors.
• This represents the first application of nonlinear MPC to a conductance-based model with limited information on unobservable states and parameters.

Conclusions:

• Data-driven nonlinear modeling combined with MPC provides a powerful tool for precise neural system control.
• This methodology overcomes limitations of traditional control approaches in neuroscience.
• Enables novel experimental designs and potential therapeutic strategies for neurological disorders.