Estimation of Functional Dependence in High-Dimensional Spike-Field Activity

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Neurotechnology

Background:

• Neural activity is encoded across diverse spatiotemporal scales, from individual neuronal spikes to large-scale local field potentials (LFPs).
• Understanding the functional relationships between spikes and LFPs is crucial for deciphering neural encoding and advancing neurotechnologies.
• Challenges include differing statistical properties and timescales of spikes and LFPs, and the high dimensionality of neural data leading to overfitting.

Purpose of the Study:

• To develop a model-based approach for estimating functional dependencies between high-dimensional LFP features and neuronal spike activity.
• To address the challenges of statistical differences, timescale disparities, and high dimensionality in neural data analysis.
• To enable more accurate prediction of neural activity and investigation of brain connectivity.

Main Methods:

• Modeled neuronal spike trains as a point process influenced by behavioral states and network LFP features.
• Employed L1-regularization for supervised and unsupervised learning to identify significant spike-LFP dependency parameters.
• Utilized the Akaike Information Criterion (AIC) to enforce model sparsity, balancing complexity and predictive power.

Main Results:

• The proposed method successfully identified functional dependencies between spikes and LFP features.
• The model significantly improved the prediction accuracy of spiking activity.
• The approach achieved better prediction with substantially fewer parameters compared to methods without regularization.

Conclusions:

• The developed model-based approach effectively estimates spike-LFP functional dependencies.
• This method offers a powerful tool for investigating brain connectivity across multiple spatiotemporal scales.
• The findings contribute to a better understanding of neural encoding and the development of improved neurotechnologies.