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). Under the assumption that spike arrivals at synapses have Poisson statistics, we derive analytically stochastic mean-field dynamics for the effective synaptic coupling between finite-size populations undergoing Tsodyks-Markram STP. The novel mean-field equations account for both finite number of synapses and correlations between the neurotransmitter release probability and the fraction of available synaptic resources. Comparisons with Monte Carlo simulations of the microscopic model show that in both feedforward and recurrent networks the mesoscopic mean-field model accurately reproduces stochastic realizations of the total synaptic input into a postsynaptic neuron and accounts for stochastic switches between Up and Down states as well as 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 efficient and mathematically tractable mean-field models.
Introduction: The Mesoscopic Gap
In computational neuroscience, researchers face a "Goldilocks" problem when modeling brain activity. Microscopic models simulate every individual neuron and synapse, offering high biological realism but at a crushing computational cost. Conversely, macroscopic rate models treat whole populations as single smooth variables, which is computationally efficient but often ignores "finite-size noise"—the inherent randomness in small groups of neurons—and the dynamic nature of synapses.
The gap lies at the mesoscopic scale, involving groups of a few hundred to a few thousand neurons, such as those in a cortical column. At this scale, individual spike randomness matters, and synapses constantly change strength through Short-Term Plasticity (STP). Previous theories often ignored STP or assumed infinite population sizes, missing crucial noise-driven dynamics.
Methods: Extending the Mean-Field Theory
The researchers extended mesoscopic population dynamics to include the Tsodyks-Markram model of STP. This model tracks two variables per synapse: Utilization ($u$), the probability of neurotransmitter release, and Resources ($x$), the fraction of available neurotransmitter.
The core innovation is a second-order moment-closure approximation. Instead of tracking $u$ and $x$ for every synapse, the authors derived five differential equations tracking the means and variances/covariances of these variables across the population. Crucially, this 2nd-order model accounts for the correlation between $u$ and $x$, which are both driven by the same incoming spikes. This allows the model to capture the finite-size noise that emerges in limited synaptic populations.
Results and Key Insights
The theory was validated against Monte Carlo simulations of microscopic spiking neural networks, using both Poisson neurons and more realistic Generalized Integrate-and-Fire (GIF) neurons.
High Accuracy and Realism
In feedforward networks, the 2nd-order model significantly outperformed traditional 1st-order models. It reduced the relative error in mean synaptic input from ~5% to less than 0.3%, and error in input variance from ~30% to roughly 4%.
Capturing Complex Dynamics
- Correlations are Essential: 1st-order models fail because they ignore the correlation between synaptic release probability and resources. These correlations are significant even in stationary states.
- Noise-Induced Behavior: In small populations, random fluctuations drive behavior. The model accurately predicts "population spikes" (sudden bursts) and the switching between active (Up) and inactive (Down) states. For instance, the 2nd-order model correctly matched "residence times" in bistable networks, whereas 1st-order models significantly overestimated the duration of the "Up" state.
- Complexity Reduction: The mesoscopic model reduces a population of $N$ neurons with $pN$ synapses to just 4 or 5 equations, offering a massive speedup for large-scale simulations.
Discussion and Future Outlook
While the 2nd-order model represents a major advance, it currently assumes Poisson spike statistics. While effective, it faces challenges with neurons showing strong spike-history effects, such as absolute refractoriness. Future work may involve using renewal processes to better model non-random firing timings and extending the theory to handle stochastic vesicle release in synapses. This mesoscopic population theory provides a mathematically tractable path for reducing detailed biophysical models into efficient cortical microcircuit simulations.
Summary updated