Brain signal predictions from multi-scale networks using a linearized framework.

Area of Science:

• Computational Neuroscience
• Neuroscience
• Biophysics

Background:

• Simulations of neural activity are crucial for interpreting experimental data and understanding neural mechanisms.
• Extracellular measurements of brain signals, reflecting transmembrane currents, are commonly used.
• Accurate prediction of low-frequency extracellular signals (≲ 300Hz) from large-scale neuronal network models is challenging due to computational limitations.

Purpose of the Study:

• To develop a computational framework for predicting spatiotemporal filter kernels of extracellular signals originating from synaptic activity.
• To account for the biophysics of neurons, neural populations, and recurrent connections in signal prediction.
• To improve the computational efficiency of biophysics-based signal predictions from large-scale neural network models.

Main Methods:

• Developed a framework to predict spatiotemporal filter kernels for extracellular signals.
• Signals are generated by convolving population spike rates with derived kernels.
• Kernels are derived using linearized synapse and membrane dynamics, cell distributions, conduction delay, and volume conductor models.

Main Results:

• The proposed framework accurately captures the spatiotemporal dynamics of extracellular signals from detailed multicompartment neuron networks.
• Persistent synapse activation necessitates accounting for changes in effective membrane time constants.
• The framework significantly reduces signal prediction times compared to biophysically detailed models.

Conclusions:

• The developed framework provides accurate, biophysics-based extracellular signal predictions for large-scale spiking and rate-based neural network models.
• This work offers insights into the approximate linear dependence of low-frequency extracellular signals on spiking activity.
• The software tool LFPykernels serves as a reference implementation.