Inferring network dynamics and neuron properties from population recordings.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Neural Dynamics

Background:

• Understanding neural computation requires characterizing emergent multiscale dynamics.
• Existing methods often rely on bottom-up or linearization approaches, limiting their scope.

Purpose of the Study:

• To propose and validate a novel model-driven identification procedure for neural networks.
• To infer network properties from response dynamics without bottom-up analysis or linearization.

Main Methods:

• Applied a model-driven identification procedure to excitatory integrate-and-fire neuron networks with spike frequency adaptation (SFA).
• Utilized a mean-field theory parameterized by identified elements, analyzing network response to brief stimulations.
• Extracted the input-output gain function and linked it to microscopic properties like SFA decay time constant and synaptic efficacy.

Main Results:

• Successfully inferred dynamic timescales and single-neuron properties from network-level responses.
• Demonstrated the ability to capture system dynamics across bifurcations and different dynamical regimes.
• Validated the method's robustness and generality through controlled simulations with good agreement between expected and identified values.

Conclusions:

• The proposed method offers a powerful, non-linear, and holistic approach to understanding neural network computation.
• It provides direct links between macroscopic network dynamics and microscopic neuronal properties.
• The methodology is generalizable and applicable to experimental preparations like cultured neuron networks and brain slices.