Modeling multiscale causal interactions between spiking and field potential signals during behavior.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Brain activity involves dynamics across multiple spatiotemporal scales, including discrete spike trains and continuous field potentials.
• Understanding neural processes requires modeling causal interactions both within and across these scales.
• Standard causality measures are insufficient due to differing data types (binary vs. continuous).

Purpose of the Study:

• Develop computational tools to recover and model multiscale neural causality during behavior.
• Assess the performance of multiscale causality modeling on neural datasets.
• Determine if multiscale causality modeling improves neural signal prediction compared to single-scale methods.

Main Methods:

• Designed a multiscale, model-based Granger-like causality method using directed information.
• Learned point-process generalized linear models to predict spike events from spike train and field potential histories.
• Learned linear Gaussian models to predict field potential signals from their own history and either spike events or latent firing rates.

Main Results:

• The method successfully revealed true multiscale causality networks in simulations, even with model mismatch.
• Models incorporating multiscale causalities in non-human primate (NHP) data improved prediction of both spike trains and field potentials.
• Latent firing rates were found to be better predictors of field potential signals than binary spike events in NHP data.

Conclusions:

• The developed multiscale causality method effectively identifies directed functional interactions across different scales of brain activity.
• This approach enhances the understanding of neural processes by integrating information from both spike trains and field potentials.
• The findings have implications for basic neuroscience research and the development of neurotechnologies.