| name | arbor-tvb-multiscale-simulation |
| description | Multi-scale brain simulation framework integrating microscopic neural dynamics (Arbor)
with macroscopic whole-brain models (The Virtual Brain/TVB). Enables bidirectional
real-time co-simulation between detailed spiking neurons and large-scale brain networks.
Use when:
- Modeling seizures at both neural and whole-brain scales
- Bridging microscopic neuron dynamics with macroscopic brain region activity
- Co-simulating detailed neuron populations within large-scale network models
- Multi-scale neuroscience research requiring both levels of detail
- Investigating neural disorder propagation mechanisms
- MPI-based neural simulator integration
|
| metadata | {"arxiv_id":"2505.16861","published":"2025-05-22","revised":"2025-12-16","authors":"Thorsten Hater, Juliette Courson, Han Lu, Sandra Diaz-Pier, Thanos Manos","doi":"10.48550/arXiv.2505.16861","tags":["multi-scale","neuroscience","brain-simulation","seizure","arbor","tvb","mpi"],"source":"arXiv:2505.16861v2 [q-bio.NC]"} |
| license | Complete terms in LICENSE.txt |
Arbor-TVB: Multi-Scale Brain Co-Simulation Framework
Overview
Arbor-TVB is a novel framework that bridges two traditionally isolated scales in computational neuroscience:
- Microscopic scale (Arbor): Detailed biophysical neuron models, single-compartment to population-level simulations
- Macroscopic scale (TVB): Whole-brain dynamics based on anatomical connectomes and mean neural activity
This integration enables studying how local neural events (like seizure onset) propagate to whole-brain phenomena.
Core Innovation
First Integration of Arbor + TVB
The framework links these simulators for real-time bidirectional interaction:
- Arbor: Generates discrete spike events from biologically realistic neurons
- TVB: Models continuous neural population activity at regional level
- Translation layer: Converts discrete spikes ↔ continuous activity via MPI intercommunicator
Key Technical Components
1. MPI Intercommunicator Architecture
Real-time bidirectional data exchange:
Arbor (Microscopic) ←→ MPI Intercommunicator ←→ TVB (Macroscopic)
(Spikes) (Translation) (Mean Activity)
Translation mechanism:
- Arbor spikes → TVB mean firing rate per region
- TVB continuous activity → Arbor population input currents
2. Modular Design
Independent model selection per scale:
- Replace any TVB node with Arbor neuron population
- Minimal effort to translate activity across simulators
- Scale-specific model configuration without global constraints
3. Seizure Case Study Implementation
Mouse Brain Connectome:
- 38 brain regions from Allen Mouse Brain Connectivity Atlas
- Anatomically informed connectivity weights
- Region-specific Arbor populations embedded in TVB network
Seizure modeling workflow:
- Define seizure-prone region (e.g., hippocampus)
- Embed detailed Arbor neuron population with seizure-inducing parameters
- Run co-simulation to observe:
- Local seizure onset dynamics
- Propagation pathways to connected regions
- Whole-brain network response patterns
Mathematical Framework
Arbor: Hodgkin-Huxley Style Neurons
Single-compartment neuron dynamics:
C_m * dV/dt = -g_L*(V - E_L) - I_ion(V, t) + I_syn(t)
Where:
- C_m: Membrane capacitance
- g_L: Leak conductance
- E_L: Leak reversal potential
- I_ion: Ion channel currents (Na, K, etc.)
- I_syn: Synaptic input from connected neurons
TVB: Neural Mass Models
Regional mean activity model:
dS(t)/dt = -S(t)/τ + (1 - S(t)) * H(S(t)) * r_in(t)
Where:
- S(t): Mean synaptic activity per region
- τ: Synaptic decay time
- H(S): Sigmoidal response function
- r_in: Input firing rate from connected regions
Translation: Spike ↔ Mean Activity
Arbor → TVB:
mean_firing_rate = spike_count / (population_size * simulation_time)
tvb_region_input = mean_firing_rate * weight_connection
TVB → Arbor:
input_current = tvb_mean_activity * scaling_factor
for neuron in arbor_population:
neuron.I_syn += input_current
Implementation Guide
Step 1: Install Arbor and TVB
pip install arbor
pip install tvb-library tvb-framework
Step 2: Configure MPI Intercommunicator
import mpi4py.MPI as MPI
class ArborTVBCoupling:
def __init__(self):
self.comm = MPI.COMM_WORLD
self.intercomm = self.comm.Split_type(MPI.COMM_TYPE_SHARED)
def send_spikes_to_tvb(self, spikes, region_id):
spike_rate = len(spikes) / self.time_window
self.intercomm.send(spike_rate, dest=region_id, tag=0)
def receive_activity_from_tvb(self, region_id):
mean_activity = self.intercomm.recv(source=region_id, tag=1)
return mean_activity
Step 3: Define Mouse Brain Model
from tvb.datatypes import connectivity
mouse_connectome = connectivity.Connectivity.from_file(
"allen_mouse_brain_connectivity.h5"
)
regions = [
"Hippocampus", "Cortex_visual", "Cortex_frontal",
"Thalamus", "Amygdala", "Striatum", ...
]
weights = mouse_connectome.weights
Step 4: Embed Arbor Population in TVB Node
import arbor
hippocampus_cells = arbor.make_cells(
morphology='pyramidal_cell.swc',
mechanisms=['hh', 'synapse'],
n_cells=1000
)
for cell in hippocampus_cells:
cell.set_parameter('gNa', 150)
cell.set_parameter('gK', 5)
tvb_model.replace_node('Hippocampus', arbor_population=hippocampus_cells)
Step 5: Run Co-Simulation
simulator = ArborTVBCoSimulator(
arbor_model=arbor_cells,
tvb_model=tvb_network,
coupling_interval=10ms
)
simulator.inject_current('Hippocampus', 5nA, duration=100ms)
simulator.run(duration=5000ms)
propagation_pathway = simulator.get_seizure_propagation()
Seizure Propagation Results
Observed Patterns
- Onset: Seizure begins in Arbor-embedded region (e.g., hippocampus)
- Propagation: Spreads via anatomical connections to neighboring regions
- Whole-brain effect: Network-level oscillations emerge from local hyperactivity
Key Insights
- Seizure propagation follows anatomical connectivity weights
- Detailed neuron dynamics affect propagation speed and pattern
- Intervention timing depends on both local and network state
Advantages over Single-Scale Models
- Biological realism: Detailed neurons + anatomical connectivity
- Causal inference: Trace seizures from origin to whole-brain effect
- Intervention testing: Simulate treatments at appropriate scale
- Computational efficiency: Detailed only where needed, mean-field elsewhere
- Modularity: Swap models per region without global redesign
Use Cases
| Task | Approach |
|---|
| Study seizure onset mechanisms | Arbor population in focal region |
| Model drug effects on networks | Modify Arbor neuron parameters |
| Test intervention strategies | Simulate stimulation timing |
| Understand propagation pathways | Trace TVB network activity |
| Validate clinical hypotheses | Compare with patient EEG/fMRI |
Technical Requirements
- MPI: For parallel intercommunicator (mpi4py)
- Arbor: >=0.6 (Python bindings)
- TVB: >=2.0 (framework and library)
- Hardware: Multi-core CPU for MPI parallelism
- Connectome data: Allen Mouse Brain Atlas (or human connectome)
Pitfalls
- MPI synchronization: Ensure both simulators run at same timestep
- Translation scaling: Spike rate ↔ mean activity conversion must match population size
- Parameter tuning: Arbor neuron params affect TVB network stability
- Computational cost: Detailed regions expensive; limit number of Arbor nodes
- Connectome accuracy: Anatomical weights critical for propagation patterns
Connection to Neuroscience
- Multi-scale modeling: Addresses long-standing scale integration challenge
- Seizure research: Clinical relevance for epilepsy studies
- Mouse brain: Translates to human with appropriate connectome
- Neural mass models: TVB foundation grounded in mean-field theory
- Biophysical detail: Arbor neurons follow Hodgkin-Huxley formalism
References
@article{hater2025arbor-tvb,
title={Arbor-TVB: A Novel Multi-Scale Co-Simulation Framework with a Case Study on Neural-Level Seizure Generation and Whole-Brain Propagation},
author={Hater, Thorsten and Courson, Juliette and Lu, Han and Diaz-Pier, Sandra and Manos, Thanos},
journal={arXiv preprint arXiv:2505.16861},
year={2025}
}
Knowledge Graph Integration: See references/knowledge-graph-schema.md for verified kg.db schema and insert patterns.
Related Tools
This skill is based on research published on arXiv:2505.16861v2 (May 2025, revised December 2025)