This repository contains the code to reproduce the results of:
"Multiscale Brain Dynamics at the Edge of Criticality"
Rabuffo, G.; Bozzo, P.; Nguyen, B.; Depannemeacker, D.; Pompili, M.; Gollo, L.; Fukai, T.; Sorrentino, P.; Dalla Porta, L. (2025).
This project introduces a multiscale, connectome-based modeling framework that unites the study of local and global brain criticality.
By tuning neural mass models to subcritical, critical, and supercritical regimes and embedding them into the empirically derived mouse connectome, we explore how local and global dynamics interact to generate experimentally observed features of brain activity.
Key contributions of this work include:
- Demonstrating that global signatures of criticality (maximal autocorrelation, avalanche scaling, 1/f spectra) emerge only when local populations are tuned near criticality and coupled within an optimal range of global coupling.
- Showing that subcritical and supercritical regimes also reproduce meaningful dynamics (e.g., oscillations, flattened spectra), suggesting that distinct brain regions may operate at different distances from criticality.
- Revealing that structural in-strength shapes spatial gradients of timescales, reversing direction between subcritical and supercritical tuning.
- Linking global criticality with improved correspondence to empirical mouse fMRI data (functional connectivity and dynamic FC).
This framework highlights how local tuning, long-range interactions, and network topology jointly shape scale-free, flexible dynamics across the connectome.
The study uses structural and functional data from the Allen Mouse Brain Atlas and resting-state fMRI from 53 control mice under light anesthesia (medetomidine–isoflurane protocol) [Grandjean, 2020].
- Connectome: tracer-based directed structural connectivity of the mouse brain (Oh et al., 2014; Melozzi et al., 2017).
- fMRI dataset: DOI:10.34973/1he1-5c70 (CC-BY 4.0).
The repository is organized into numbered Jupyter notebooks. Running them in order reproduces all simulations and figures from the paper.
| Notebook | Description |
|---|---|
| 1) Empirical_data_processing | Preprocesses structural and functional datasets (Allen connectome, rsfMRI). |
| 2) Simulate_Local_NMM | Simulates isolated neural mass models in subcritical, critical, and supercritical regimes. |
| 3) Local_phase_space | Explores phase space structure, stability, and bifurcations of the local model. |
| 4) Two_coupled_populations | Examines how coupling shifts two regions toward or away from criticality depending on initial state. |
| 5) Whole_brain_simulations | Embeds local models into the empirical connectome and runs large-scale simulations across coupling strengths (G). |
| 6) Whole_brain_RAW_analysis | Analyzes raw neural activity: autocorrelations, metastability, avalanche statistics, timescale gradients. |
| 7) Whole_brain_BOLD_analysis | Transforms neural activity into BOLD signals (Balloon–Windkessel model), computes FC/dFC, and compares with empirical fMRI. |
- The pipeline runs end-to-end: from empirical preprocessing (1) to BOLD-level analysis and data comparison.
- Each notebook generates the figures corresponding to its stage of analysis.
- Avalanche analyses, power spectra, and autocorrelation timescales are reproduced in analysis notebooks.
- Python 3.9+
- Jupyter Notebook
- The Virtual Brain
- NumPy, SciPy, Pandas, Matplotlib, NetworkX
If you use this code, please cite:
Rabuffo, G.; Bozzo, P.; Nguyen, B.; Depannemeacker, D.; Pompili, M.; Gollo, L.; Fukai, T.; Sorrentino, P.; Dalla Porta, L. (2025). Multiscale Brain Dynamics at the Edge of Criticality.