Unsupervised functional neurocartography of the Aplysia buccal ganglion

A major limitation of large-scale neuronal recordings is the difficulty in locating the same neuron in different subjects, referred to as the "correspondence" issue. This issue stems, at least in part, from the lack of a unique feature that unequivocally identifies each neuron. One promising approach to this problem is the functional neurocartography framework developed by Frady et al. (2016), in which neurons are identified by a semi-supervised machine learning algorithm using a combination of multiple selected features. Here, the framework was adapted to the buccal ganglia of Aplysia. Multiple features were derived from neuronal activity during motor pattern generation, responses to peripheral nerve stimulation, and the spatial properties of each cell. The feature set was optimized based on its potential usefulness in discriminating neurons from each other, and then used to match putatively homologous neurons across subjects with the functional neurocartography software. A matching method was developed based on a cyclic matching algorithm that allows for unsupervised extraction of groups of neurons, thereby enhancing scalability of the analysis. Cyclic matching was also used to automate the selection of high-quality matches, which allowed for unsupervised implementation of the machine learning algorithm. This study paves the way for investigating the roles of both well-characterized and previously uncharacterized neurons in Aplysia, as well as helps to adapt this framework to other systems.