Bayesian inference of a spectral graph model for brain oscillations

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biophysics

Background:

• Brain functional and structural connectivity are key research areas, often modeled mathematically.
• The spectral graph model (SGM) offers a biophysically interpretable, parameter-efficient approach to model brain oscillations.
• Current SGM parameter estimation relies on computationally intensive annealing algorithms yielding only point estimates without confidence intervals.

Purpose of the Study:

• To develop a Bayesian inference framework for SGM parameters using simulation-based inference (SBI).
• To reduce the computational cost of SGM parameter estimation.
• To enable the quantification of parameter uncertainty and correlations.

Main Methods:

• Incorporation of simulation-based inference (SBI) into a Bayesian framework for SGM parameter estimation.
• Application of the SBI-SGM framework to resting-state magnetoencephalography (MEG) data.
• Comparison of SBI-SGM performance against traditional annealing algorithms.

Main Results:

• The SBI-SGM framework demonstrates comparable performance to annealing in recovering power spectra and alpha band spatial distribution.
• SBI significantly reduces the computational burden for parameter inference.
• The Bayesian approach allows for the analysis of parameter correlations and uncertainty via posterior distributions.

Conclusions:

• The proposed SBI-SGM framework provides a robust, efficient, and informative method for inferring SGM parameters.
• This approach enhances understanding of biophysical parameter interactions and their uncertainties.
• Simulation-based Bayesian inference holds promise for advancing generative models in clinical neuroscience applications.