Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Neural Engineering

Neural Engineering - Poster Session E

Poster Z14 - Predicting nonstationary spike-spike correlations with local field potentials

Saturday, October 26, 2024
10:00 AM - 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  Recordings of neural signals often include high frequency, action potentials (spikes) from individual neurons as well as slower population field potentials. Local field potentials (LFPs) are correlated with neural spiking activity, and previous statistical techniques have shown how stationary spike-spike correlations can be predicted by the spike-LFP relationships of individual neurons. However, both spike-spike correlations and LFPs themselves are nonstationary. Spike-spike correlations are often used to understand the representation of sensory or motor variables in the brain, the function of neural circuits, and their dysfunction during diseases, such as epilepsy and Parkinson’s. Here we develop new methods to track the nonstationary characteristics of spike-spike correlation and show that they can be predicted by LFP.

Materials and

Methods:  We develop statistical methods for fitting spike-spike correlation from LFP using a bilinear modeling framework. Using this framework, we estimate fixed spike-LFP relationships for each neuron that predict the time-varying spike-spike relationship in short time windows as a result of time-varying LFP features. We examine several LFP representations, including a wavelet filter bank and a generalized phase estimation method. We then show how this modeling approach can be extended to allow multiple LFP channels to be modeled simultaneously. We evaluate our models using large-scale multi-electrode recordings in the mouse hippocampus and visual cortex from the Allen Institute – Neuropixels dataset. Additionally, we compare our new methods to a previous approach based on a single-channel Fourier representation. In the Fourier model, the spike-spike relationships are predicted based on the spike-triggered average of the LFP for both neurons. The multi-channel filter bank model is designed to extract features from LFP across different frequency bands or scales defined by the wavelet filters. The generalized phase method aims to use the phase characteristics of LFP signals filtered within a specified frequency range using a bandpass filter, as the features. These features are used as predictors of the spike-spike correlations.

Results, Conclusions, and Discussions: We find that our models are able to accurately track both stationary and time-varying spike-spike correlations in experimental recordings. Our approach out-performs the previous single-channel Fourier method. Additionally, we find that multi-channel models are more accurate than single-channel models, and LFP frequency bands contribute differentially to predicting spike-spike correlations. Although both cortical and hippocampal spike-spike correlations vary over the recording as brain state changes, these changes are, at least partially, predicted by models that assume a fixed spike-LFP relationship. In analyses of neural coding, spike-spike correlations can be used as features to better decode sensory variables. This work may help to partition spike-spike correlations and disentangle the contributions of external variables and internally generated, ongoing oscillatory activity. This may lead to improvement in brain-machine interfaces and may help to understand the role that correlation plays in neural circuit function and disease.

Acknowledgements (Optional):