Embedding optimization reveals long-lasting history dependence in neural spiking activity

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Information Theory

Background:

• Neural spiking patterns reflect information processing.
• Efficient coding and temporal integration leave distinct statistical footprints.
• Classical methods struggle to disentangle history dependence strength and timescale.

Purpose of the Study:

• To develop a novel information-theoretic approach to quantify neural spiking history dependence.
• To disentangle the strength and timescale of history dependence in single neuron spiking.
• To apply and validate this method on model and experimental neural data.

Main Methods:

• Quantified history dependence using mutual information between past and current spiking.
• Developed an embedding-optimization approach for estimating mutual information from limited experimental data.
• Applied the method to extracellular spike recordings from hippocampus, visual cortex, and in vitro preparations.

Main Results:

• The novel method successfully disentangles history dependence strength and timescale in model neurons.
• Experimental data revealed independent variation of history dependence strength and timescale across preparations.
• Hippocampus showed strong, long history dependence; visual cortex showed weak, short dependence; in vitro showed strong, short dependence.

Conclusions:

• This information-theoretic approach provides a more comprehensive characterization of neural history dependence.
• The findings highlight distinct information processing strategies across different neural circuits and preparations.
• The developed toolbox and guidelines facilitate the application of this novel method in neuroscience research.