Estimation of neuronal firing rate using Bayesian Adaptive Kernel Smoother (BAKS).

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Neurons communicate information using action potentials (spikes).
• Neuronal firing rate is typically estimated by averaging across multiple trials.
• Single-trial firing rate estimation is crucial for understanding neural encoding but remains challenging.

Purpose of the Study:

• To develop a novel, accurate method for estimating neuronal firing rate from single trials.
• To address limitations of current state-of-the-art single-trial firing rate estimation techniques.

Main Methods:

• Developed the Bayesian Adaptive Kernel Smoother (BAKS) method.
• Utilized a kernel smoothing technique with adaptive bandwidth selection via an empirical Bayesian framework.
• Employed Gaussian kernel functions and analytically derived kernel bandwidths.

Main Results:

• BAKS demonstrated superior performance compared to established methods (OKS, VKS, Locfit, BARS) on synthetic and real spike train data.
• Evaluated performance using synthetic data from inhomogeneous Gamma (IG) and inhomogeneous inverse Gaussian (IIG) models.
• Applied BAKS to real spike train data from non-human primate motor and visual cortex.

Conclusions:

• The proposed BAKS method offers a significant improvement for single-trial firing rate estimation.
• BAKS can enhance understanding of neural encoding mechanisms in cognitive tasks.
• Potential to improve the performance of brain-machine interface (BMI) decoders.