Statistical analysis of synaptic connectivity in neural networks

An analytical examination is presented of methods for estimating interneuronal connectivities. The analysis is based on a mathematical model of neural networks, where the dynamics and nonlinearity of each neuron are explicitly represented, and the spike generation is modeled as a doubly stochastic point process on the half-line. The normalized joint peri-stimulus-time scatter diagram is shown to be the most accurate and versatile representation of the neural connectivities. The normalization procedure for removing extraneous correlations due to stimulus effects is presented and compared favorably to the shuffling method. The method is readily extended from pairwise correlations to the case of an arbitrary number of neural recordings, and an analytical evaluation is made of the information gained with the increased number of recorded neurons.<>