Abstract
With the availability of cellular-resolution connectivity maps, connectomes, from the mammalian nervous system, it is in question how informative such massive connectomic data can be for the distinction of local circuit models in the mammalian cerebral cortex. Here, we investigated whether cellular-resolution connectomic data can in principle allow model discrimination for local circuit modules in layer 4 of mouse primary somatosensory cortex. We used approximate Bayesian model selection based on a set of simple connectome statistics to compute the posterior probability over proposed models given a to-be-measured connectome. We find that the distinction of the investigated local cortical models is faithfully possible based on purely structural connectomic data with an accuracy of more than 90%, and that such distinction is stable against substantial errors in the connectome measurement. Furthermore, mapping a fraction of only 10% of the local connectome is sufficient for connectome-based model distinction under realistic experimental constraints. Together, these results show for a concrete local circuit example that connectomic data allows model selection in the cerebral cortex and define the experimental strategy for obtaining such connectomic data.
Introduction
A central debate in modern neuroscience concerns the utility of cellular-resolution "wiring diagrams," or connectomes. While advancements in electron microscopy allow us to map every neuron and synapse within small tissue volumes, critics argue these datasets risk becoming uninterpretable "data graveyards." The fundamental question is whether structural data alone can serve as an arbiter between competing theoretical models of brain function. For any given cortical process, multiple models—ranging from random networks to highly structured circuits—can often explain the same functional output. This study tests whether connectomic data can definitively identify which of these models is implemented in the mammalian cortex.
Methodological Framework
The researchers employed a sophisticated statistical approach known as Approximate Bayesian Computation with Sequential Monte Carlo (ABC-SMC). This framework allows for model selection by comparing simulated connectomes to observed (or to-be-measured) data using a set of summary statistics.
Instead of comparing every individual synapse, the complex connectomes were reduced to six simple graph statistics:
- Reciprocity: The frequency of bidirectional links within and across excitatory and inhibitory populations.
- Recurrency: The prevalence of specific motifs, such as 5-neuron cycles.
- In-Out Degree Correlation: The relationship between a neuron's incoming and outgoing connection counts.
By simulating thousands of connectomes from candidate models, the ABC-SMC algorithm calculates the posterior probability, determining which model most likely generated the observed structural pattern.
Biological Constraints and Circuit Modeling
The study focused on the "barrel" cortex, a local circuit module in Layer 4 of the mouse primary somatosensory (S1) cortex. This circuit typically contains approximately 2,000 neurons, with a 90/10 split between excitatory (ExN) and inhibitory (IN) neurons.
Seven candidate models were evaluated, including random networks (ER-ESN), distance-dependent networks (EXP-LSM), and models derived from plasticity rules (STDP-SORN). These models were constrained by real-world biological data, such as pairwise connectivity probabilities and reciprocity ranges, to ensure biological plausibility.
Key Results
High-Accuracy Model Discrimination
Under ideal conditions, the framework identified the correct circuit model with over 90% accuracy. This demonstrates that structural data is indeed highly informative for distinguishing between different organizational principles of the cortex.
The "10% Rule" and Experimental Feasibility
One of the most significant findings is that a complete connectome is not required for accurate model selection. Mapping just 10% of the neurons in a local circuit provides sufficient statistical power to identify the underlying model. This finding has profound implications for experimental design, potentially reducing the labor required for reconstruction by an order of magnitude (e.g., from 180,000 work hours to 18,000).
Robustness to Noise
The Bayesian approach proved resilient to experimental errors. The method maintained high accuracy even when 15-25% of synapses were missing or incorrectly identified, provided these error rates were accounted for in the statistical priors.
Structural Fingerprints of Function
When Recurrent Neural Networks (RNNs) were trained on different tasks, such as texture classification or sequence memorization, they developed unique structural signatures. These "fingerprints" were most prominent in the strongest connections, suggesting that the network's functional identity is embedded in its structural backbone.
Discussion and Future Directions
While the success of this method relies on the assumption of population-level homogeneity in wiring rules, it provides a powerful tool for falsifying incorrect biological hypotheses. Future work may incorporate additional structural details, such as synaptic weights (correlated with spine volume), to distinguish between models that differ primarily in their plasticity rules. Ultimately, this research defines a clear experimental strategy for using connectomics to resolve long-standing questions about cortical architecture.
Reference: Klinger, E., Motta, A., Marr, C. et al. Cellular connectomes as arbiters of local circuit models in the cerebral cortex. Nat Commun 12, 2785 (2021). https://doi.org/10.1038/s41467-021-22856-z
Summary updated