Robust Estimation of Sparse Narrowband Spectra from Neuronal Spiking Data

Area of Science:

• Computational Neuroscience
• Signal Processing

Background:

• Spectral analysis is crucial for understanding neural processes but challenging for spiking data due to its binary nature.
• Existing methods struggle to accurately capture the spectral properties of neuronal spiking.

Purpose of the Study:

• To develop a robust method for estimating the power spectral density (PSD) of the neural covariate driving neuronal population spiking.
• To enable spectral analysis of binary neural recordings.

Main Methods:

• Utilized a Bernoulli spiking model with a logistic map of a stationary process.
• Employed sparsity-promoting priors for maximum a posteriori estimation of PSD.
• Developed an efficient posterior sampling procedure for confidence intervals.

Main Results:

• The proposed method significantly outperforms existing techniques in extracting frequency content from spiking data.
• Applied to clinical data, the estimated PSD closely matched the local field potential (LFP) signal.
• Results support the role of LFP as a key neural covariate in rhythmic cortical activity.

Conclusions:

• The developed technique robustly analyzes the harmonic structure of spiking activity.
• It operates independently of LFPs and without prior assumptions on spectral content.
• The method is applicable to various binary datasets beyond neuroscience, such as heart rate data.