New York University, Abu Dhabi
Courant Institute, NYU
November 22, 2016
Spike Triggered Regression on Neuronal Network Reconstruction
In recent years, technological advances in neuroscience, such as calcium imaging, multi-electrode array recording, make it possible to record firing activity of neuronal populations with single cell resolution. Meanwhile, intracellular recording technique, which can be applied only to a small number of neurons simultaneously, is able to provide detailed information of neuronal voltage traces. We are interested in how to combine information of population firing activities and the detailed voltage traces to unravel the coupling structure of the underlying neuronal network. We propose a spike-triggered regression (STR) method to address this question. Theoretically, we show that it can well capture the response of the subthreshold voltage trace to presynaptic spikes and we can deploy STR to accurately infer the coupling between neurons. Under the conditions that (1) the network dynamics is nearly synchronous or (2) recording time is short (10~100s), many other reconstruction methods fail to reconstruct the network connectivity. We can demonstrate that the integrate-and-fire neuronal network connectivity can be faithfully reconstructed even under those two conditions.