A probabilistic framework for synapse localization and class discovery in the mouse whisker barrel cortex

A thorough understanding of the brain's diverse molecular architecture and complex neural circuitry is critical to uncovering the mechanisms that underlie higher-level information processing. Synapses are an integral component of neural circuits because they propagate signals from one neuron to the next. However, because of the astounding multitude and diversity of synapses in the brain, a set of high-resolution, high-precision tools is necessary for extracting enough information from a tissue sample to recover its synaptic structure. To this end, many state-of- the-art molecular imaging methods, including array tomography, have recently focused their efforts on building large-scale datasets annotated with synaptic tissue markers. Manual processing of these images, however, is too time-intensive to be practical. Here, we present a computational method for synapse detection and classification that aims to locate and characterize all synapses in a mouse barrel cortex. Our technique uses a nonparametric Bayesian network to represent the underlying biological process in order to allow for flexibility in the total numbers of synapses and
synapse types that are found. Using this data-driven approach, we are able to detect subtle but significant patterns with implications for neuroscience research.