According to PubMed, a January 2026 review by Hashemi, Jirsa, and colleagues in IEEE Reviews in Biomedical Engineering introduces Virtual Brain Twins (VBT), a framework for creating personalized digital replicas of individual brains. The technology integrates structural and functional brain data into computational models that can predict treatment responses and optimize clinical interventions.

The Population Trial Problem

Current clinical methods rely on population-wide trials that measure average treatment effects across large groups. A drug that works for 60% of patients may fail or cause harm in the remaining 40%. This approach overlooks individual variability in brain structure, connectivity patterns, and neural dynamics.

Mechanism-based trials, which test interventions based on specific neural circuit properties, remain rare in neuroscience. The brain’s complexity makes it difficult to predict how a given intervention will affect a particular patient’s neural networks. Virtual Brain Twins address this gap by creating patient-specific models that can simulate treatment effects before clinical application.

Technical Architecture

VBT construction proceeds through four stages. First, researchers acquire structural brain data using diffusion tensor imaging (DTI) to map white matter connectivity between brain regions. This produces a patient-specific anatomical scaffold showing how different neural populations are physically connected.

Second, the anatomical data is coupled with mathematical models of neuronal population dynamics. Rather than simulating individual neurons, VBT uses mesoscopic models that represent the average activity of thousands of neurons in a given brain region. These models capture key dynamic properties like excitability, oscillation frequencies, and response to perturbations.

Third, the coupled brain model is simulated to generate predictions about brain-wide activity patterns. The simulation can be run under different conditions, such as varying drug concentrations, stimulation parameters, or lesion locations. This allows researchers to test hypotheses about how specific interventions will affect the patient’s brain dynamics.

Fourth, Bayesian inference algorithms tune the model parameters to match the patient’s actual brain activity as measured by functional MRI or electroencephalography. This calibration step ensures that the virtual twin accurately reflects the individual’s neural dynamics rather than generic brain properties.

Technology Readiness Level: TRL 4-5. VBT has progressed from laboratory prototypes to validation in clinical cohorts, but standardized clinical workflows and regulatory approval pathways are still under development.

Clinical Applications Demonstrated

The review documents VBT applications across multiple neurological conditions. In epilepsy research, patient-specific models have been used to predict seizure onset zones and test the effects of different surgical resection strategies before the actual procedure. This capability is particularly valuable for drug-resistant epilepsy cases where surgical intervention is being considered.

For multiple sclerosis patients, VBT models incorporate lesion locations and white matter damage patterns to simulate how disease progression affects brain network function. Researchers can test whether specific disease-modifying therapies are likely to slow functional decline based on the patient’s connectivity profile. This approach mirrors concepts in our analysis of whole brain emulation feasibility, where understanding individual neural architecture becomes critical for accurate simulation.

In healthy aging studies, VBT has been applied to understand how normal age-related changes in connectivity affect cognitive function. These models help distinguish healthy aging patterns from early neurodegenerative changes, supporting earlier intervention for conditions like Alzheimer’s disease.

The review also discusses extending VBT to Parkinson’s disease, where models could predict optimal deep brain stimulation parameters for individual patients. Current stimulation protocols use standardized targeting coordinates, but patient-specific models could account for individual differences in basal ganglia connectivity.

Network Modeling Framework

The core innovation in VBT is the network modeling approach that couples mesoscopic neural population models through patient-specific anatomical connectivity. Each brain region is represented by a neural mass model that captures population-level dynamics, such as local field potentials and firing rates.

These regional models are then coupled according to the patient’s white matter connectivity as measured by DTI. The coupling strength and conduction delays between regions depend on the number, myelination, and length of connecting axon bundles. This architecture allows the model to simulate how activity patterns propagate through the patient’s specific neural network.

The framework uses nonlinear dynamical systems theory to analyze how the brain operates near critical transitions between different activity states. Bifurcation analysis identifies parameter ranges where small changes in neural excitability or connectivity can trigger large-scale state changes, such as the transition from normal activity to seizure dynamics.

Consciousness Research Applications

The review identifies consciousness research as an emerging application area for VBT. Different theories of consciousness make specific predictions about which brain regions and connectivity patterns are necessary for conscious experience. Patient-specific models could test whether disrupting particular networks abolishes consciousness in ways predicted by specific theories.

This application connects to recent work on transcranial focused ultrasound for consciousness mapping, where precise stimulation of deep brain structures can reveal causal relationships between neural activity and conscious perception. VBT models could predict which brain regions are most likely to show consciousness-relevant effects when stimulated, guiding experimental design.

For whole brain emulation research, understanding the neural substrate of consciousness is fundamental. If consciousness depends on specific network architectures or temporal dynamics, successful brain emulation must preserve these features. VBT provides a framework for testing which neural properties are sufficient for conscious experience.

Brain-Computer Interface Integration

The review discusses VBT applications in brain-computer interface (BCI) development. Patient-specific models can predict which neural signals provide the most reliable control information for a given individual. This capability could accelerate BCI training by identifying optimal electrode placements and decoding algorithms before invasive implantation.

For motor BCIs, VBT models simulate how different motor cortex lesion patterns affect the remaining neural signals available for decoding. This information guides decisions about whether a patient is a good candidate for a motor BCI and which movements can be reliably decoded from their specific neural architecture. Recent clinical trials, such as BCI rehabilitation for multiple sclerosis patients, demonstrate the therapeutic potential of personalized BCI approaches.

Sensory feedback BCIs could also benefit from VBT modeling. By simulating how stimulation of different somatosensory areas produces conscious percepts, researchers could design feedback patterns that feel natural to the patient. This application requires accurate models of both sensory encoding and conscious perception.

Computational Requirements

Running VBT simulations requires substantial computational resources. A whole-brain model with hundreds of coupled neural mass models must be integrated over time using small time steps to capture relevant dynamics. Simulating one hour of brain activity can take several hours of computation on modern workstations.

The review notes that cloud computing infrastructure makes VBT more accessible to clinical sites that lack high-performance computing resources. Researchers can upload patient imaging data to secure cloud platforms where the model construction, simulation, and analysis occur remotely. Results are then returned to the clinical team for interpretation.

Parameter estimation using Bayesian inference is particularly computationally intensive. The algorithm must explore a high-dimensional parameter space to find values that best match the patient’s observed brain activity. Recent developments in gradient-based optimization and machine learning acceleration have reduced inference times from days to hours for typical clinical applications.

Regulatory and Ethical Considerations

The review emphasizes several barriers to widespread VBT adoption. Regulatory agencies have not established clear pathways for approving computational models as medical decision support tools. Questions remain about liability when a model-based prediction turns out to be incorrect and leads to suboptimal treatment decisions.

Data privacy concerns arise from the use of patient-specific brain imaging data. Brain connectivity patterns contain identifying information that could potentially be used to infer cognitive or psychological traits. Secure data handling protocols and patient consent procedures must be developed to protect this sensitive information.

Clinical validation remains the primary scientific challenge. VBT predictions must be demonstrated to improve patient outcomes compared to standard care in randomized controlled trials. Such trials are expensive and time-consuming, but necessary to establish clinical utility and justify reimbursement from healthcare payers.

Relation to Bridge Protocol Research

For the Bridge Protocol and whole brain emulation research, VBT represents an intermediate milestone on the path to complete brain simulation. The technology demonstrates that personalized brain models can capture clinically relevant neural dynamics using current imaging and computational methods.

The framework also highlights remaining gaps in our understanding of brain function. VBT models work at the level of neural populations, averaging over thousands of individual neurons. Whether this mesoscopic scale captures sufficient detail for consciousness preservation during emulation remains an open question, as discussed in our review of the State of Brain Emulation Report 2025.

The success of VBT in predicting treatment responses suggests that certain aspects of brain function depend more on network architecture than on fine-grained cellular details. This observation has implications for determining the necessary resolution for whole brain emulation. If population-level dynamics are sufficient for some cognitive functions, emulation might not require simulating every synapse.

Path Forward

VBT technology represents a convergence of advances in brain imaging, computational modeling, and high-performance computing. The framework provides interpretable predictions about individual brain function that can guide clinical decision-making in neurology and psychiatry.

As clinical validation studies accumulate and regulatory pathways clarify, VBT could become a standard tool in personalized medicine for brain disorders. The technology also serves as a research platform for testing theories about brain function, consciousness, and the neural basis of cognition.

For those working toward whole brain emulation, VBT demonstrates both the promise and limitations of current brain modeling capabilities. The technology shows that patient-specific simulations are technically feasible and clinically useful, while also highlighting the substantial work remaining to achieve complete functional emulation of human cognition.

Official Sources

According to PubMed, this article reviews research from:

Primary Paper: Hashemi, M., Depannemaecker, D., Saggio, M., Triebkorn, P., Rabuffo, G., Fousek, J., Ziaeemehr, A., Sip, V., Athanasiadis, A., Breyton, M., Woodman, M., Wang, H., Petkoski, S., Sorrentino, P., & Jirsa, V. (2026). Principles and Operation of Virtual Brain Twins. IEEE Reviews in Biomedical Engineering, 19, 111-139. DOI: 10.1109/RBME.2025.3562951

Author Affiliations: Primary authors from Institut de Neurosciences des Systèmes, Aix-Marseille Université, Marseille, France

Related Resources:

• PubMed Record
• IEEE Reviews in Biomedical Engineering