A grouped beta process model for multivariate resting-state EEG microstate analysis on twins.

Area of Science:

• Neuroscience
• Computational Biology
• Statistical Modeling

Background:

• Electroencephalography (EEG) microstate analysis examines discrete temporal segments of brain electrical activity.
• Brain activity exhibits stable periods within microstates, interspersed with rapid, non-random transitions between states.

Purpose of the Study:

• To develop a Bayesian nonparametric model for simultaneously estimating the number and dynamics of EEG microstates.
• To analyze resting-state EEG data from twin pairs to understand within-pair similarity and individual differences in brain activity patterns.

Main Methods:

• Utilized a Markov switching vector autoregressive (VAR) framework integrating a hidden Markov model (HMM) for state dynamics and a VAR model for within-state activity.
• Applied an Indian buffet process to model an infinite library of microstates, allowing participants to select a finite subset, thus capturing twin similarity.
• Fit the model at the twin-pair level, sharing information across participants and epochs to enforce within-participant and within-twin-pair similarity.

Main Results:

• The Bayesian nonparametric model successfully identified a sparse set of EEG microstates.
• The model effectively captured within-twin-pair similarity in microstate selection and switching dynamics.
• Demonstrated the ability to differentiate between highly similar and dissimilar twins based on their selected microstate spaces.

Conclusions:

• The proposed model offers a robust framework for EEG microstate analysis, adept at handling complex brain dynamics.
• This approach enhances the understanding of neurophysiological similarities and differences within families, particularly twins.
• The method provides a data-driven way to define and analyze the fundamental building blocks of brain electrical activity.