Deriving connectivity from spiking activity in detailed models of large-scale cortical microcircuits

Inferring detailed cortical microcircuit connectivity is essential for uncovering how information is processed in the brain. A common method in vivo uses short-lag spike cross-correlations to derive putative monosynaptic connections, but inactive neurons and correlated firing can hinder the derivation accuracy. Previous computational studies that developed methods to derive connectivity from cross-correlations employed simplified or small network models and thus did not address the above key confounds of physiological large-scale networks. We tested connectivity derivation using simulated ground-truth spiking from detailed models of human cortical microcircuits in different layers and between key neuron types. While derivation accuracy was high for cortical layer 5 microcircuits, we showed that low firing and inactive neurons in layer 2/3 microcircuits resulted in poor performance. We then showed that general activation paradigms for layer 2/3 microcircuits led to only a moderate improvement in derivation performance, due to a trade-off between reducing the proportion of inactive neurons and increasing correlated overactive neurons. We further improved the connection derivation performance using a more refined activation paradigm leading to jittered moderate spiking, which decreased inactive neurons without incurring unwanted correlations. Our results address key physiological challenges and provide methods to improve performance in deriving connections from spiking activity in large-scale neuronal microcircuits.