Global search metaheuristics for neural mass model calibration

Area of Science:

• Computational neuroscience
• Dynamical systems theory
• Neuroimaging analysis

Background:

• Neural mass models (NMMs) are crucial for understanding brain circuitry and dynamics in research.
• Model calibration, or fitting, infers NMM parameters from data to reveal physiological mechanisms.
• NMM calibration is challenging due to non-identifiability, high dimensionality, and nonlinearity.

Purpose of the Study:

• To investigate the limitations of existing NMM calibration methods, like dynamic causal modelling (DCM).
• To motivate and evaluate calibration methods that explore broader parameter spaces.
• To compare the performance of approximate Bayesian computation (ABC) and evolutionary search metaheuristics (ESMs) for NMM calibration.

Main Methods:

• Utilized dynamical systems theory to analyze NMM calibration complexities.
• Applied approximate Bayesian computation (ABC) and evolutionary search metaheuristics (ESMs).
• Tested methods on recreating electroencephalographic (EEG) recordings during eyes-closed resting state.

Main Results:

• Evolutionary search metaheuristics (ESMs) demonstrated superior computational efficiency and accuracy compared to ABC.
• ESMs effectively mapped feasible parameter sets for NMMs to replicate EEG data.
• ESMs showed greater robustness against biases from complex cost landscapes inherent in NMM calibration.

Conclusions:

• ESMs are more effective than ABC for NMM calibration, particularly for complex biological systems.
• The findings support the integration of ESMs into future efforts for modeling brain dynamics.
• Improved calibration methods enhance confidence in using NMMs for interpreting neuroimaging data.