Bayesian modelling and analysis of spatio-temporal neuronal networks
Fabio Rigat, Mathisca de Gunst and Jaap van Pelt
This paper illustrates a novel dynamic Bayesian network for stochastic integrate-and-fire neuronal systems. We adopt a Bayesian hierarchical perspective to introduce at different model stages the parameters characterizing the neuronal spiking process over a discrete time grid, the unknown structure of functional connectivities and its dependence
on the spatial arrangement of the neurons. We compute estimates for all the model parameters and predictions for future spiking states through the standard Gibbs sampler using a shrinkage prior. The paper includes the analyses of a simulated dataset and of experimental in
vitro multi-electrode spike trains. In the latter case, we find that the estimates of the neuronal parameters are consistent with their biological interpretation. Furthermore, estimation of the network structure reveals a complex pattern of functional relationships which
significantly depend on the spatial distribution of the
neurons. Evaluation of the goodness of fit and of the prediction residuals indicates that the model can adequately explain the observed spiking patterns.