functional-whole-brain-models

name: functional-whole-brain-models description: > Functional Whole-Brain Models (FWBM) methodology bridging bottom-up whole-brain modeling and top-down neuroconnectionism. Combines biophysically detailed simulations with functional-performance-driven deep neural networks. Use when: designing brain-scale computational models, integrating structure and function in neural modeling, building neuroconnectionist models with biological grounding, developing hybrid brain models that achieve both biological fidelity and functional competence. Activation: whole-brain modeling, neuroconnectionism, functional brain model, brain simulation, WBM, FWBM, biophysically detailed brain, brain DNN integration, brain foundation model, brain dynamics modeling. Based on arXiv:2605.18118 (May 2026).

Functional Whole-Brain Models (FWBM)

Core Concept

Proposes a new paradigm that bridges two prominent computational neuroscience traditions:

• Bottom-up Whole-Brain Modeling (WBM): Biophysically detailed simulations of brain structure and dynamics. Achieves biological fidelity but lacks functional competence.
• Top-down Neuroconnectionism: Deep neural networks optimized for functional performance. Achieves task competence but limited biological grounding.

FWBM unifies both by incorporating anatomical constraints and biophysical realism into architectures that are also trained for functional competence.

Key Contributions (arXiv:2605.18118)

1. Hybrid Architecture: Combines structural connectivity matrices (from DWI/tractography) with deep learning architectures that can be trained end-to-end on cognitive tasks.
2. Biological Constraints as Regularizers: Uses empirical brain data (fMRI, MEG, DWI) as architectural and training constraints rather than mere validation targets.
3. Multi-scale Integration: Bridges micro-scale (neuron-level dynamics) with macro-scale (region-level functional connectivity) through hierarchical modeling.
4. Functional Validation: Models must simultaneously reproduce neural activity patterns AND achieve behavioral task performance.

Methodology Framework

Step 1: Structural Foundation

• Load individual or template structural connectome (SC matrix)
• Define regional parcellation (e.g., Schaefer, AAL, HCP-MMP)
• Map SC to network connectivity weights

Step 2: Biophysical Embedding

• Incorporate neural mass models or mean-field approximations at each node
• Use empirical delays from tractography-derived fiber lengths
• Set coupling strengths based on empirical SC weights

Step 3: Functional Training

• Define target tasks (cognitive, perceptual, motor)
• Train with task loss + biological regularization loss
• Biological loss terms: FC matching, spectral matching, dynamic FC matching

Step 4: Validation

• Compare simulated FC to empirical FC
• Compare temporal dynamics (power spectra, metastability)
• Evaluate behavioral task performance

Implementation Patterns

# Conceptual framework
class FunctionalWholeBrainModel:
    def __init__(self, sc_matrix, regions, delays):
        self.sc = sc_matrix  # structural connectivity
        self.regions = regions  # parcellation
        self.delays = delays  # conduction delays
        
    def forward(self, input_signal, params):
        # Biophysical dynamics at each node
        # Coupled through structural connectivity
        # Differentiable for gradient-based training
        pass
    
    def loss(self, predictions, targets, empirical_data):
        task_loss = compute_task_loss(predictions, targets)
        bio_loss = compute_biological_loss(predictions, empirical_data)
        return task_loss + lambda_ * bio_loss

Activation Keywords

• whole-brain modeling
• neuroconnectionism
• functional brain model
• brain simulation
• brain DNN
• biophysical brain model
• structure-function coupling
• brain foundation model
• computational brain modeling
• neural mass modeling

Related Skills

• brain-dit-fmri-foundation-model
• neural-dynamics-universal-translator
• brain-inspired-snn-pattern-analysis
• computational-neuroscience-in-llm-era

References

• Paper: arXiv:2605.18118 (May 2026)
• Related: The Virtual Brain platform, Dynamic Causal Modeling, Neural Mass Models