Extracting cell types from neuronal wiring diagrams


March 27, 2015 - 11:00am
NW 353
About the Speaker
Gad Minz (Hebrew University)

The field of connectomics promises to quantify the connectivity between all neurons within a tissue volume, such as a single barrel in rodent somatosensory cortex or a local patch of retina. Given the inevitable variability of the local circuit from one region to a nearby one (e.g., variability across different cortical columns), and across time and individuals, an important goal of connectomics is to extract from a small copies of a wiring diagrams the underlying statistical connectivity rules. Furthermore, these rules may reveal insight into the structure-function relation of the circuit.

I will describe work aimed at extracting information about cell types from neuronal wiring diagrams. Here we classify cells into types by the statistics of their connectivity to other cells; this classification may or may not coincide with traditional morphological and molecular-genetic classification.

We use the Stochastic Block Model (SBM) on directed graphs as a generative model of a local cortical circuit. We use an approximate Bayesian inference method to extract from an observed wiring diagram, the statistical connectivity rules, as well as the type of each cell in the circuit. We apply the method to a local connectome of the rat barrel cortex, estimated from detailed 3D reconstructions of cell morphologies and their locations. Our work highlights some of the key factors that determine the ability to extract cell types and other connectivity rules from dense connectomes.