Arbor-TVB: A Novel Multi-Scale Co-Simulation Framework with a Case Study on Neural-Level Seizure Generation and Whole-Brain Propagation

Computational neuroscience has traditionally focused on isolated scales, limiting understanding of brain function across multiple levels. While microscopic models capture biophysical details of neurons, macroscopic models describe large-scale network dynamics. Integrating these scales, however, remains a significant challenge. In this study, we present a novel co-simulation framework that bridges these levels by integrating the neural simulator Arbor with The Virtual Brain (TVB) platform. Arbor enables detailed simulations from single-compartment neurons to populations of such cells, while TVB models whole-brain dynamics based on anatomical features and the mean neural activity of a brain region. By linking these simulators for the first time, we provide an example of how to model and investigate the onset of seizures in specific areas and their propagation to the whole brain. This framework employs an MPI intercommunicator for real-time bidirectional interaction, translating between discrete spikes from Arbor and continuous TVB activity. Its fully modular design enables independent model selection for each scale, requiring minimal effort to translate activity across simulators. The novel Arbor-TVB co-simulator allows replacement of TVB nodes with biologically realistic neuron populations, offering insights into seizure propagation and potential intervention strategies. The integration of Arbor and TVB marks a significant advancement in multi-scale modeling, providing a comprehensive computational framework for studying neural disorders and optimizing treatments.

💡 Research Summary

This paper introduces a novel multi‑scale co‑simulation framework that tightly couples the microscopic neural simulator Arbor with the macroscopic whole‑brain platform The Virtual Brain (TVB). The authors address the long‑standing challenge of bridging the vast gap between detailed single‑neuron dynamics (microseconds) and large‑scale network activity (seconds to minutes) by employing an MPI intercommunicator that enables real‑time, bidirectional data exchange. Arbor, written in C++ with a high‑level Python interface, solves the cable equation for morphologically detailed neurons and can scale to thousands of cells using GPU acceleration. TVB, on the other hand, builds whole‑brain models from MRI/DTI data, representing each brain region as a mean‑field variable (e.g., firing rate or average membrane potential).

The integration works as follows: spikes generated by Arbor are collected, time‑stamped, and transformed into a continuous activity signal that serves as an input to the corresponding TVB node. Conversely, the continuous TVB output is monitored; when it exceeds a user‑defined threshold, a current injection is generated and delivered to the Arbor network, effectively turning the macroscopic signal back into spikes. Synchronisation between the two simulators is achieved at each integration step using MPI barriers, ensuring that both sides advance with consistent time steps despite their intrinsic scale differences.

A key design principle is modularity. The framework allows any TVB node to be replaced by an Arbor‑based neuronal population (or vice‑versa) with minimal code changes. The translation layer is exposed as customizable Python functions, enabling researchers to implement alternative mappings such as voltage‑to‑firing‑rate functions, conductance‑based synaptic updates, or more biologically realistic coupling schemes.

To demonstrate the system, the authors use a mouse brain connectome consisting of 90 regions. One region is swapped out for an Arbor network of detailed neurons based on an extended Hodgkin‑Huxley model (Depannemaecker et al., 2022). This model includes intracellular and extracellular dynamics of K⁺, Na⁺, and Cl⁻, a sodium‑potassium pump, and an external potassium bath (K_bath). By varying K_bath from 9 mM (resting) to 17 mM, the network reproduces a spectrum of behaviors: tonic spiking, bursting, seizure‑like events (SLE), sustained ictal activity, and depolarization block. In the co‑simulation, when the Arbor sub‑network enters an SLE state, the resulting high‑frequency spike train is translated into an elevated TVB activity variable, which then propagates through the structural connectome to other regions, inducing synchronized pathological activity at the macroscopic level.

Performance measurements show that the combined system runs roughly twice as fast as a pure CPU implementation, thanks to Arbor’s GPU acceleration, while memory consumption remains modest. The authors note that the current implementation is optimised for swapping a single TVB node; extending to multiple simultaneous replacements will increase communication overhead and requires further optimisation. Additionally, the spike‑to‑continuous conversion relies on threshold parameters that must be calibrated against experimental data to ensure physiological relevance.

Future work outlined includes (1) scaling to multi‑node replacements and assessing scalability on large supercomputing clusters, (2) incorporating slower plasticity mechanisms (e.g., synaptic weight changes, homeostatic regulation) to study long‑term dynamics, (3) validating the framework with human patient data (EEG/MEG/fMRI) for personalized digital twin applications, and (4) testing virtual therapeutic interventions such as deep brain stimulation or pharmacological modulation within the co‑simulation environment.

In summary, the Arbor‑TVB co‑simulator represents the first fully integrated platform that unites biophysically detailed neuronal models with whole‑brain dynamics. Its bidirectional, real‑time coupling, modular architecture, and demonstrated ability to model seizure initiation and propagation make it a powerful tool for neuroscientists seeking to explore cross‑scale mechanisms of brain disorders and to design and evaluate novel therapeutic strategies.