Interaction graph estimation for the first olfactory relay of an insect

We consider the model of interacting neurons proposed in [8] and use the statistical selection procedure proposed in [4] to infer the interaction graph from observed neural activity (spike trains) of the first olfactory relay of an insect. We propose a pruning method to deal with small sample sizes. We use simulations to test the efficiency of the estimator and to fix the parameters involved in its construction. We then use these parameters to analyze actual data recorded from the locust (Schistocerca americana) antennal lobe. In order to test the performance of the proposed estimator on our spike train data, the data set is split in two parts and the resulting connection graphs are compared.