Source-linked AI summary
The scientific case for brain simulations
Gaute T. Einevoll, Alain Destexhe, Markus Diesmann, Sonja Grün, Viktor Jirsa, Marc de Kamps, Michele Migliore, Torbjørn V. Ness, Hans E. Plesser, Felix Schürmann
TL;DR
The paper addresses how neuroscience can bridge poorly understood links between neurons, networks, and system-level brain function. It argues for general-purpose simulators spanning biological detail levels and predicting multiple experimental modalities. These simulators would support systematic comparison and refinement of candidate network models, although incomplete network data and computational access remain important boundaries.
Problem
Network mechanisms linking neurons to brain-wide function remain poorly understood, and mechanistic models of specific biological brain networks are still in their infancy.
Method
The paper proposes community-developed, general-purpose simulators that support multiple neuron-model resolutions and predict electrical, magnetic, and optical signals from candidate network models.
Results
The paper concludes that multimodal simulators are likely indispensable tools for bridging neuron- and system-level neuroscience and comparing candidate models with experiments.
Takeaways & Limitations
Brain simulators should serve as mathematical observatories for testing many candidate hypotheses rather than embodying a single hypothesis about brain function.
Takeaways & Limitations
Initial network models remain plausible skeletons because key data, including synaptic strength and plasticity, are lacking, while simulations can require inaccessible supercomputing resources.
Abstract
from arXiv · showhide
A key element of the European Union's Human Brain Project (HBP) and other large-scale brain research projects is simulation of large-scale model networks of neurons. Here we argue why such simulations will likely be indispensable for bridging the scales between the neuron and system levels in the brain, and a set of brain simulators based on neuron models at different levels of biological detail should thus be developed. To allow for systematic refinement of candidate network models by comparison with experiments, the simulations should be multimodal in the sense that they should not only predict action potentials, but also electric, magnetic, and optical signals measured at the population and system levels.
1 Introduction
Brain research has characterized individual neurons more successfully than the networks they form, while mechanistic models of specific biological brain networks remain immature. Large-scale simulations are therefore presented as necessary tools for bridging neuron- and system-level understanding.
- Network behavior remains poorly understood despite substantial research across molecular, cellular, circuit, and system levels.
- Descriptive receptive-field models provide limited insight into how neuronal networks generate observed representations.
- Real brain networks contain heterogeneous neural populations and structured synaptic connections, unlike many stylized mechanistic models.
- Mechanistic modeling of biological neural networks that mimic specific brains or brain areas is still in its infancy.
- Modern supercomputers are making simulations of networks containing hundreds of thousands or millions of neurons feasible.
- The article argues for multipurpose brain simulators whose long-term development requires community efforts rather than individual researchers or groups.
2 Brain simulations
Brain simulations should connect biological scales and neuron-model resolutions while predicting experimentally measurable signals. The proposed simulators should be multimodal, reusable across models, and capable of supporting comparisons across multiple measurement modalities.
- Brain function spans interconnected spatial scales from atoms and molecules to whole organisms, making scale bridging central to understanding.
- Large-scale projects primarily link neuron-level mechanisms to networks because network behavior is difficult to understand through qualitative reasoning alone.
- Large-scale models use morphologically detailed, spatially extended, point-neuron, and firing-rate representations of neural activity.
- Brain simulations should predict action potentials together with population-level LFP, ECoG, and VSDI signals and system-level EEG and MEG signals.
- A barrel column can be modeled with multicompartment, integrate-and-fire, or population firing-rate units, while the simulator predicts multiple signals from shared network activity.
- The paper distinguishes equations and parameters in a model from the software simulator that executes it and the resulting simulation.
- HBP simulators are designed to execute many different brain-network models rather than being tied to one fixed model.
3 Network simulators not tailored to specific brain-function hypotheses
The paper argues that brain simulators should function as general tools for testing candidate network hypotheses, not as implementations of any single hypothesis. Their usefulness is bounded by incomplete knowledge of network organization and missing data needed to specify models.
- Brain simulators should test many existing and future hypotheses rather than being designed around one specific hypothesis about brain function.
- The Newton analogy separates a tool for generating testable predictions from the hypotheses that the tool evaluates.
- Calculus enabled comparison of predictions from gravitational hypotheses with measurements, illustrating the proposed role of brain simulators.
- No generally accepted theory currently explains how networks of millions or billions of neurons generate salient brain functions.
- Candidate network models require anatomical, electrophysiological, spatial, and connectivity information that is not yet fully available.
- Initial large-scale models are therefore plausible skeleton models used to guide simulator construction and further exploration.
- Brain simulators should compute experimentally relevant predictions from candidate network models so model merit can be assessed against experiments.
4 Use of brain network simulators
Brain network simulators are proposed as tools for testing mechanistic candidate models against experiments and refining them through multimodal predictions. They can also provide virtual ground-truth data for validating experimental-analysis methods and let researchers explore model behavior across biological scales, but computational demands limit access.
- 4.1 Biological imitation game: Simulators can test candidate network models against experiments, progressively narrowing attention toward models whose predictions best mimic recordings.The paper frames leading models as hypotheses that remain open to challenge by new experiments and candidate models.
- 4.1 Biological imitation game: Model selection remains unsettled because simulations may be compared using spike timing, firing rates, LFP or VSDI features, spectral measures, or behavior.The appropriate criterion depends partly on assumptions about whether information is encoded in firing rates or detailed spike timing.
- 4.2 Validation of data-analysis methods: Simulations can generate benchmarking data with known neuron and network activity for validating methods that analyze spikes, synfire chains, current-source densities, and functional connectivity.Such virtual data support testing population- and systems-level analyses against known ground truth.
- 4.3 Use by wider research community: Researchers could use candidate models to predict stimulus-evoked spiking, pharmacological perturbation effects, LFP contributions, and EEG differences between models.These questions span cellular perturbations, population signals, and systems-level measurements.
- 4.3 Use by wider research community: Large simulations may require supercomputers, limiting access for researchers without suitable infrastructure or experience maintaining complex software.Web-based services are proposed as one way to provide remote access to centralized supercomputing resources.
5 Discussion and outlook
The paper presents brain simulators as mathematical observatories for testing candidate hypotheses and argues that a coordinated, multi-resolution simulation infrastructure is needed to bridge brain scales. It also outlines extensions toward larger networks, longer timescales, multimodal biology, stimulation, and broader modeling domains.
- Brain simulators should function as mathematical observatories for testing candidate hypotheses, rather than embodying a single hypothesis themselves.The paper likens simulators to tools for imaging brain structure or activity.
- Different simulator resolutions are needed: detailed multicompartmental models expose dendritic integration, point neurons enable larger networks, and population models simplify computation and concepts.These levels trade biological detail and parameter complexity against computational scale and ease of fitting.
- Population-level neural-mass models should eventually be derived from individual-neuron network models or fitted to experiments rather than remaining largely phenomenological.
- Future infrastructure should support larger networks, longer processes such as plasticity and homeostasis, extracellular and glial interactions, and electrical or magnetic stimulation.The authors connect these extensions to studying learning and stimulation effects.
- The paper focuses on bottom-up neuron-network simulators and does not address subcellular molecular-signaling or molecular-dynamics simulators.It also notes that network models can represent cognitive processes such as attention, language, decision making, and learning.
- Understanding brain function remains limited until spatial and temporal scales are connected, making mathematical modeling and simulation central to scale bridging.The paper compares this challenge with numerical weather prediction and smartphone engineering.