Estimating the interaction graph of stochastic neuronal dynamics by observing only pairs of neurons

TL;DR

This paper introduces an efficient algorithm to identify neuronal interactions from pairwise spiking activity, capable of handling large data sets with proven accuracy bounds and consistency in detecting true interactions.

Contribution

It presents a novel, computationally efficient method for inferring neuronal interaction graphs from pairwise observations, with theoretical guarantees.

Findings

Algorithm effectively detects interactions with low false positive rates

Proven bounds on detection error probabilities

Algorithm is consistent in identifying true neuronal connections

Abstract

We address the questions of identifying pairs of interacting neurons from the observation of their spiking activity. The neuronal network is modeled by a system of interacting point processes with memory of variable length. The influence of a neuron on another can be either excitatory or inhibitory. To identify the existence and the nature of an interaction we propose an algorithm based only on the observation of joint activity of the two neurons in successive time slots. This reduces the amount of computation and storage required to run the algorithm, thereby making the algorithm suitable for the analysis of real neuronal data sets. We obtain computable upper bounds for the probabilities of false positive and false negative detection. As a corollary we prove the consistency of the identification algorithm.