A Model for Single Neuron Activity With Refractory Effects and Spike Rate Estimation Techniques.

Area of Science:

• Computational Neuroscience
• Statistical Signal Processing
• Neural Encoding

Background:

• Neural spike trains are often modeled using Poisson point processes for rate estimation.
• The refractory phenomenon in neurons violates the Poisson assumption due to induced history dependency.
• Accurate estimation of time-varying firing rates is crucial for understanding neural dynamics.

Purpose of the Study:

• To develop a Maximum Likelihood (ML) estimation framework for time-varying neuronal firing rates that incorporates history dependencies.
• To present a novel neural spiking model capable of representing absolute and relative refractory effects.
• To improve the accuracy and goodness of fit for firing rate estimation in neural data.

Main Methods:

• Developed a self-exciting point process model based on an exponential of polynomial excitation function.
• Employed non-convex optimization and model order selection techniques for ML estimator derivation.
• Validated the framework using simulated data with refractory periods and real neuronal recordings.

Main Results:

• The proposed ML estimation technique, accounting for the complete refractory phenomenon, demonstrated improved accuracy on simulated data.
• Application to measured neuronal data showed enhanced goodness of fit compared to methods ignoring refractory effects.
• The new framework outperformed other commonly used firing rate estimation techniques.

Conclusions:

• The developed ML framework effectively models neural spike trains with refractory properties.
• This approach offers a more accurate and robust method for estimating time-varying neuronal firing rates.
• The findings have implications for advancing neural data analysis and understanding neural coding.