Region-specific brain models reveal distinct cerebellar dynamics

Introduction

Large-scale brain simulations have advanced significantly, most use simplified mathematical models that treat all brain regions identically. A one-size-fits-all approach seems to fall short, particularly for the cerebellum, as it misses unique properties of specialised circuits. 

In this study, published in npj Systems Biology and Applications, researchers at the University of Pavia and IRCCS Mondino Foundation built region-specific mean-field models to describe how different parts of the cerebellum generate collective neural dynamics. By combining biological realism with computational efficiency of virtual brain models, the approach helps bridge microscopic neural processes (microcircuitry) and whole-network behaviour. 

Research question and rationale

Standard models — known as neural mass models — collapse all neurons in a region into two populations: excitatory and inhibitory. While computationally efficient, this ignores the distinctive microcircuit properties of individual brain regions. The present study investigates whether incorporating region-specific microcircuit properties into mean-field models (1) can improve their ability to reproduce realistic cerebellar dynamics, support more accurate large-scale brain simulations, also beyond the cerebellum itself.

The cerebellum offers a strong test case. It contains more than half of all neurons in the human brain and features a highly regular and structured microcircuit with four distinct cell types. Its regions differ in cellular composition, connectivity, and function. These differences shape how neural activity emerges and propagates, with specific excitation and inhibition patterns. However, standard neural mass models do not reflect these specific features. Moreover, MRI tractography cannot reliably resolve the cerebellum's internal connectivity, leaving key connections unmapped.

Methods and approach - From microcircuit models to virtual brain simulations

Using a multiscale approach, the team developed a cerebellar mean field (CRBL MF)model that preserves physiological properties such as different cell types, synaptic interactions, and excitation-inhibition balance of four principal neuronal populations while remaining computationally efficient. Starting from detailed experimental models of individual cerebellar neurons (microcircuitry), they constructed spiking neural networks representing micrometric tissue portions of the cerebellum. Working with MRI data from eight Human Connectome (2) Project participants, they built subject-specific brain networks, and integrated this cerebellar model into The Virtual Brain (TVB) platform, creating a cerebellar mean field virtual brain (cMF-TVB).

Key results: Improved fit to empirical brain activity

Model performance was assessed by comparing simulated resting-state BOLD (3) brain-activity signals with empirical functional fMRI recordings.  Importantly, the researchers curated intra-cerebellar connectivity by incorporating anatomical knowledge that cannot be directly captured by MRI tractography. 

The study shows that cerebellar regions do not behave identically when described at the population level. Region-specific mean-field models produced distinct activity patterns, reflecting underlying biological differences that generic models fail to capture. The cerebellum represented by the region-specific cMF-TVB reduced prediction errors by approximately 50% compared to standard neural mass models - a statistically significant improvement across all eight subjects and all cerebellar cortical regions. Importantly, benefits extended beyond the cerebellum itself, improving simulations of deep cerebellar nuclei and whole-brain signals, demonstrating that a single region-specific model can have far-reaching effects on global brain dynamics — without any parameter optimisation.

What is new is not the use of mean-field models themselves, but their systematic adaptation to distinct cerebellar regions within a unified modelling framework. The cMF-TVB model also reproduced physiologically plausible oscillations across multiple frequency bands. Because each neuronal population is tracked separately, simulated signals can be linked to specific cell types - something no previous large-scale simulation has achieved for the cerebellum.

Significance and implications for digital brain modelling

By showing that region-specific mean-field models capture cerebellar dynamics more accurately than generic neuron mass models, this study strengthens the foundations of whole-brain modelling -a priority for the EBRAINS research infrastructure - and advances the creation of accurate digital twins of patients' brains (4) that could transform neurological diagnosis and treatment. 

Although the present study focuses on healthy participants, the ability to identify which neuronal populations generate specific brain rhythms has direct clinical relevance. In fact, abnormal cerebellar oscillations appear in Parkinson's disease, schizophrenia and dementia. The cMF-TVB provides a mechanistic framework for interrogating how changes in specific neuronal populations propagate into whole-brain dysfunction, and could ultimately inform patient-specific treatment strategies.

Future work will extend the approach to additional brain regions, pathological conditions and full connectome-based simulations, where EBRAINS computational infrastructure can play a key enabling role  to extract individual biophysical parameters for personalised brain models.

Read the paper

Region-specific mean field models enhance simulations of local and global brain dynamics

Lorenzi, R.M., Palesi, F., Casellato, C. et al. Region-specific mean field models enhance simulations of local and global brain dynamics. 

npj Syst Biol Appl 11, 66 (2025); doi: https://doi.org/10.1038/s41540-025-00543-9

Mini Glossary

(1)Mean-field model 

A mathematical framework describing the average activity of a large neuronal population, rather than tracking each cell individually.

(2) Connectome

A map of structural connections between brain regions, derived from diffusion-weighted MRI data.

(3) BOLD signal

Blood Oxygen Level Dependent signal — the measure of local brain activity captured by functional MRI, based on changes in blood oxygenation.

(4) Digital brain twin

A personalised computational model of an individual's brain, built from their imaging data, used to simulate function and explore disease mechanisms.

This research was supported by the EBRAINS research infrastructure, Horizon Europe Virtual Brain Twin Project (GA No. 101137289), EBRAINS 2.0 (GA No. 101147319), and the Italian PNRR programmes MNESYS and EBRAINS-Italy.