Toward Personalized Neuroscience: Evaluating Individual-Level Information in Neural Mass Models.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Macroscale brain modeling with neural mass models (NMMs) simulates whole-brain dynamics.
• Connectome-based NMMs enable personalized models but struggle with individual neural characteristics due to small sample sizes and computational limits.

Purpose of the Study:

• To develop and apply an efficient, algorithmically differentiable NMM for large-scale datasets.
• To investigate the capacity of NMM parameters to capture individual-specific neural characteristics and behavioral metrics.

Main Methods:

• Utilized a differentiable reduced Wong-Wang (RWW) model for efficient optimization.
• Applied the model to resting-state fMRI data from 1444 participants.
• Optimized models with varying parameter complexities (4, 658, 23,875) and assessed variance explanation in functional connectivity (FC).

Main Results:

• Optimized RWW models explained 4%, 19%, and 56% of empirical FC variance with increasing parameter complexity.
• Subject identification accuracy based on simulated FC improved from <1% to nearly 100% with increased complexity.
• Individual-level correlations between model parameters and attributes (age, gender, IQ) were minimal (effect sizes: η² ≤ 0.03, β ≤ 0.234).

Conclusions:

• Current RWW NMM implementations robustly replicate resting-state dynamics but lack granularity for individual behavioral metrics.
• A critical alignment problem exists: neural patterns and behavioral constructs may not map one-to-one.
• Future research requires more expressive models and multimodal data to bridge the gap between neural similarity and clinical personalization.