EBRAINS researchers launch Virtual Brain Inference tool to advance personalised brain modelling through Bayesian inference

Developed by researchers at Aix-Marseille University, led by Prof. Viktor Jirsa, VBI uses advanced machine learning and Bayesian statistical inference (2) to automatically estimate the model parameters that best explain recorded brain data - bridging a major gap in personalised brain modelling.

1. The context

A main goal in neuroscience is to understand the brain’s complex dynamics and what mechanisms determine cognition, brain function, and brain diseases. Network neuroscience contributes to this by creating whole-brain network models, called virtual brain modelling, that  combine computer simulations of brain dynamics with individual brain imaging data (like MRI or eEEG) to simulate both healthy and pathological brain activity, like in epilepsy, Parkinson’s, Alzheimer’s, stroke, and mental health disorders that alter brain function.

They can also be personalised using patient-specific brain scans to distinguish between healthy from abnormal brain activities, and potentially guide future treatments.

2. Bridging the Gap in Brain Simulation: the Inverse Modelling

Tools like The Virtual Brain (TVB) have simplified simulating brain activity –  a process known as the forward problem. However a key challenge remains unsolved: the lack of automated accessible and efficient tools that can work in the opposite direction - the inverse problem, which involves starting from the outcome, i.e. the empirical brain scan data showing abnormal activity, to infer the hidden patient-specific biological mechanisms causing it.

It's hard to automatically fine-tune the set of control parameters (3) of whole-brain brain models that best explain recorded brain activity recorded - especially when using different types of brain scans and imaging techniques that vary in detail and timing.

3. The solution – Virtual Brain Inference (VBI)

The newly released Virtual Brain Inference (VBI) software package directly addresses this inverse problem. This open-source software package, openly available on EBRAINS, uses machine learning and Bayesian inference - a probabilistic reasoning method - to automate the estimation of brain model parameters from different neuroimaging modalities. Key features include:

• Fast simulations on different devices (CPUs and GPUs)

• Efficient handling of various data types, including (s)EEG, MEG, and fMRI

• Probabilistic, interpretable results  determined using statistical methods that include quantified uncertainty, via Bayesian inference

• Application of smart machine learning techniques to improve accuracy

Through computer-based experiments (called in-silico testing), the scientists demonstrated that VBI can reliably and accurately work with commonly used virtual brain models and their associated brain scan data - making whole-brain simulations more predictive, scalable and clinically relevant.

4. Towards precision neuroscience

Freely accessible via the EBRAINS platform, VBI enables researchers worldwide to build more precise brain simulations, better predict the causes of complex brain disorders, and their responses to treatment through personalised brain modellling – using both non-invasive and invasive brain recordings.

Its applications can advance network neuroscience, enhance our understanding of brain function and disorders and ultimately support the development of precision medicine for conditions such as epilepsy, Alzheimer’s,and Parkinson’s.

Glossary of technical terms

(1) Whole-brain or virtual brain models: Computational models informed by personalised anatomical data, i.e., a set of equations describing regional brain dynamics placed at each node, which are then connected through structural connectivity matrix. These models focus on the average activity of groups of neurons and connect them using information from brain scans that show the brain's wiring (called the connectome).

(2) Bayesian Rule: A fundamental belief updating principle that calculates the probability of a hypothesis – previously based on incomplete information - when new evidence becomes available.

(3) Control (generative) parameters: The bifurcation parameters, setting, or configuration within a generative model that controls the synthesis of data and potentially represents causal relationships.

Original publication (preprint):

Abolfazl Ziaeemehr, Marmaduke Woodman, Lia Domide, Spase Petkoski, Viktor Jirsa, Meysam Hashemi, Virtual Brain Inference (VBI): A flexible and integrative toolkit for efficient probabilistic inference on virtual brain models, eLife, 2025,
https://doi.org/10.7554/eLife.106194.1