Classifying dynamic transitions in high dimensional neural mass models: A random forest approach.

Area of Science:

• Computational neuroscience
• Systems neuroscience
• Mathematical biology

Background:

• Neural mass models (NMMs) are vital for understanding brain rhythms but face challenges in high-dimensional parameter space exploration.
• Classical methods like numerical continuation are insufficient for characterizing NMM dynamics across numerous parameters.
• Efficiently linking NMM parameters to emergent dynamics is crucial for both health and disease research.

Purpose of the Study:

• To develop a statistical framework for efficiently exploring the relationship between NMM parameters and their emergent dynamics.
• To enable the characterization of dynamics in high-dimensional parameter spaces of NMMs.
• To assess the relative importance of various parameters in influencing NMM dynamics.

Main Methods:

• A statistical framework combining tree-based methods and random forests was developed.
• Simulations were used to create a database, transforming the mathematical model into a data-driven resource.
• Random forests partitioned parameter space based on dynamic features, enabling rapid exploration and identification of transitions.

Main Results:

• The framework efficiently explores high-dimensional parameter spaces of NMMs.
• It accurately identifies approximate locations of qualitative dynamic transitions.
• Analysis of a seizure dynamics model highlighted the critical role of the inhibitory sub-system and previously overlooked parameters.

Conclusions:

• The introduced statistical framework offers an efficient method for exploring NMM parameter spaces.
• It aids in understanding the mechanisms underlying brain rhythms and transitions to pathological states like seizures.
• This approach facilitates more efficient, person-specific calibration of NMMs by constraining high-dimensional parameter spaces.