Efficiency and reliability in biological neural network architectures

Neurons in a neural circuit exhibit astonishing diversity in terms of the numbers and targets of their synaptic connections and the statistics of their spiking activity. We hypothesize that this diversity is the result of an underlying tension in the neural code between reliability – highly correlated activity across trials on the single neuron level – and efficiency – highly uncorrelated activity between neurons within a trial. Specifically, certain architectures of connectivity foster efficient activity while others foster the opposite, i.e., robust activity. Both coexist in a neural circuit, leading to the observed long-tailed and highly diverse distributions of connectivity and activity metrics, and allowing the robust subpopulations to promote the reliability of the network as a whole.

To test this hypothesis we developed a notion of the complexity of the connectivity of a subpopulation and used it to analyze several openly available connectomes, revealing that they all exhibited wide complexity distributions. Using co-registered functional data and simulations of a morphologically detailed network model, we found that low complexity sub-networks were indeed characterized by efficient spiking activity, and high complexity subnetworks by reliable but inefficient activity. Moreover, for neurons in cortical input layers, the focus was on increasing reliability and for output layers on increasing efficiency. To progress from describing correlations to establishing causation, we manipulated the connectivity in a biologically realistic model and showed that complex subnetworks indeed promote the reliability of the network as a whole. Our results improve our understanding of the neural code, demonstrating that the code itself is as diverse as the neuronal connectivity and activity, and must be understood in the context of the efficiency/reliability tradeoff.