A01Inferring networks from neuronal signals and predicting emergent activity patterns

 Advanced recording techniques started to provide us with a huge number of parallel spike trains. We shall focus on the following two objectives, aiming at developing methods of analyzing parallel data and their application.

(A) Reconstructing neuronal circuitry from parallel spike trains
 Cortical neurons influence the firing of other neurons through synaptic connections. Accordingly, it may be possible to infer the inter-neuronal connection from neuronal spike co-occurrences. However, connectivity cannot be corroborated by a small number of the instances, because there are a number of unobserved neurons that may have also contributed for the spiking, and we have to wait long enough until the frequency of spike co-occurrences becomes statistically significant. We develop a formula for assessing the necessary duration of observation for confirming the presence of functional connections. The formula could be useful for designing a new experiment, such that how long neuronal activity should be recorded for reconstructing a neuronal circuitry on the demanded level of connectivity strength.

(B) Inferring neuronal activity that may emerge on a given network
 Animal’s sensation and behavior are based on the cooperative activity of multitude of neurons in the brain, but their details are still obscure. Neuronal activity may fluctuate largely when animals are behaving or stimulated extrinsically, but it may also fluctuate even in the absence of stimulus, as in the default network activity of the brain at a resting state. We have recently found the condition that self-exciting systems exhibit spontaneous fluctuations (Onaga and Shinomoto, Sci Rep 2016). We are going to develop a method of inferring the potential fluctuating activity, given partial knowledge of the neuronal connectivity.