Machine learning and topology classify neuronal morphologies

The shapes of neuronal morphologies are essential for the dynamical properties of the brain, as their branching patterns govern different functional properties of cell types. However, disagreements on the definition of neuronal classes due to the subjective views of the experts have spawned several efforts to find an objective way of deriving a morphology classification. We combine machine learning with a variety of mathematical tools to group neuronal morphologies into stable classes. We show that different methods perform optimally on different use-cases. Thus, we propose a combination of traditional and novel techniques as an optimal toolkit to explore classification of rodent neurons into robust groups. Based on these methods we present a robust classification of both inhibitory and excitatory cell types in the rodent somatosensory cortex.