Presented by Nima Soltani, PhD student, Stanford University

Abstract:
An important problem in experimental neuroscience is one of connectivity: when recording from two neurons, can we infer whether or not there is a synaptic connection between them? To make this inference, previous works employed directed information as a measure of Granger causality, which in this case is the notion that if the knowledge of one neuron's firing pattern can help us better predict the firing pattern of the other, then a causal connection exists. While this method is a useful one, there are situations where the conventional estimate of the directed information between a pair of neurons can result in incorrect conclusions. These problems can be overcome by revisiting the underlying assumptions of Markovity in the firing patterns and making use of side information that is available to the experimenter, allowing for a more physiologically relevant estimate of the directed information. Using results from compartmental model simulations of neurons, the benefits of using these constraints over the original directed information calculation will be shown, and future extensions will then be discussed.

Background reading:
C.J. Quinn, T.P. Coleman, N. Kiyavash, and N.G. Hatsopoulos, "Estimating the directed information to infer causal relationships in ensemble neural spike train recordings", Journal of computational neuroscience, vol. 30, no. 1, pp. 17-44, 2010.