Mesoscopic population equations for spiking neural networks with synaptic short-term plasticity

Valentin Schmutz, Wulfram Gerstner, Tilo Schwalger

Abstract

Coarse-graining microscopic models of biological neural networks to obtain mesoscopic models of neural activities is an essential step towards multi-scale models of the brain. Here, we extend a recent theory for mesoscopic population dynamics with static synapses to the case of dynamic synapses exhibiting short-term plasticity (STP). The extended theory offers an approximate mean-field dynamics for the synaptic input currents arising from populations of spiking neurons and synapses undergoing Tsodyks-Markram STP. The approximate mean-field dynamics accounts for both finite number of synapses and correlation between the two synaptic variables of the model (utilization and available resources) and its numerical implementation is simple. Comparisons with Monte Carlo simulations of the microscopic model show that in both feedforward and recurrent networks, the mesoscopic mean-field model accurately reproduces the first- and second-order statistics of the total synaptic input into a postsynaptic neuron and accounts for stochastic switches between Up and Down states and for population spikes.

Introduction: Bridging the Scale Gap

In computational neuroscience, a persistent "scale gap" exists between microscopic models, which simulate individual neurons and synapses with high biophysical realism but at great computational cost, and macroscopic models, which describe the average activity of large populations but often ignore the critical "noise" and biological details driving neural computation.

Mesoscopic models aim to bridge this gap by describing the collective dynamics of populations of neurons (typically 100–1,000) found in cortical columns. A critical missing piece in this middle ground has been the inclusion of Synaptic Short-Term Plasticity (STP). Unlike static synapses, biological synapses facilitate (strengthen) or depress (weaken) based on recent activity.

Comparison of microscopic (B) and mesoscopic (C) modeling of synaptic input from a presynaptic population (A).

The Challenge of Synaptic Plasticity

The standard framework for modeling STP is the Tsodyks–Markram model, which tracks two variables: utilization ($u$, the fraction of resources used) and available resources ($x$). While theories existed for STP in infinite-sized populations, they often failed to account for the "finite-size noise" inherent in actual biological networks. Without accounting for this noise and the correlations between synaptic variables, mesoscopic models cannot accurately replicate collective phenomena like spontaneous bursts of activity or state transitions.

A Second-Order Mean-Field Approach

The authors developed a new mathematical framework that translates the complex, microscopic behavior of individual synapses into efficient, population-level equations. The key innovation is a second-order approximation that moves beyond simply tracking average values.

  1. Accounting for Correlations: The model derives five coupled ordinary differential equations (ODEs) that track not just the means of the synaptic variables ($u$ and $x$), but also their variances and the covariance between them. Because the dynamics of $u$ and $x$ are non-linearly coupled, their correlation directly shifts the average behavior.
  2. Gaussian Approximation: At each spike time, the joint distribution of the synaptic variables is approximated as a bivariate Gaussian, allowing for a tractable mathematical description.
  3. Stochastic Mapping: A randomization assumption is used to map population-level spikes to effective "mean-field" synapses, eliminating the need to track every individual neuron's firing history.

Validation and Key Insights

The mesoscopic equations were validated using Monte Carlo simulations of microscopic spiking neural networks (Poisson or Generalized Integrate-and-Fire neurons) as the ground truth benchmark.

  • Statistical Accuracy: The second-order model accurately reproduces the first- and second-order statistics (mean and variance) of synaptic inputs. In contrast, first-order models (which ignore correlations) consistently underestimate the coefficient of variation by up to 30% and miscalculate mean synaptic input during transients.
  • Noise as a Driver: In small populations, finite-size fluctuations are not merely noise; they are the drivers of population spikes and switches between Up and Down states. The mesoscopic model successfully replicates the irregular timing of these events in small networks ($N=100$) and correctly predicts their disappearance as population size increases ($N=5000$).
  • Computational Efficiency: For a population of $N$ neurons with recurrent connections, a microscopic model requires $2pN$ equations (where $p$ is connection probability). The mesoscopic model requires only 4 additional equations per population, regardless of $N$, offering a massive speedup for large-scale brain simulations.

Limitations and Future Directions

While highly effective, the model relies on a "Poisson-like" assumption for spike statistics. This is less accurate for neurons with very strong spike-history effects, where the agreement becomes qualitative. Furthermore, the current model uses a deterministic version of STP, whereas real biological vesicle release is highly stochastic. Extending this framework to include stochastic release and more complex plasticity models remains a promising area for future research.

This extended mesoscopic population theory provides a mathematically tractable and numerically efficient path for reducing detailed biophysical models of cortical microcircuits into large-scale simulations of brain activity.

Summary updated

Abstract

Coarse-graining microscopic models of biological neural networks to obtain mesoscopic models of neural activities is an essential step towards multi-scale models of the brain. Here, we extend a recent theory for mesoscopic population dynamics with static synapses to the case of dynamic synapses exhibiting short-term plasticity (STP). The extended theory offers an approximate mean-field dynamics for the synaptic input currents arising from populations of spiking neurons and synapses undergoing Tsodyks-Markram STP. The approximate mean-field dynamics accounts for both finite number of synapses and correlation between the two synaptic variables of the model (utilization and available resources) and its numerical implementation is simple. Comparisons with Monte Carlo simulations of the microscopic model show that in both feedforward and recurrent networks, the mesoscopic mean-field model accurately reproduces the first- and second-order statistics of the total synaptic input into a postsynaptic neuron and accounts for stochastic switches between Up and Down states and for population spikes. The extended mesoscopic population theory of spiking neural networks with STP may be useful for a systematic reduction of detailed biophysical models of cortical microcircuits to numerically efficient and mathematically tractable mean-field models.
BibTeX
@article{schmutz2020mesoscopic,
  author = {Valentin Schmutz and Wulfram Gerstner and Tilo Schwalger},
  title = {{Mesoscopic population equations for spiking neural networks with synaptic short-term plasticity}},
  journal = {The Journal of Mathematical Neuroscience},
  year = {2020},
  doi = {10.1186/s13408-020-00082-z},
}
Figure