Abstract
AbstractMotivationRecent advances in reconstructing 3D neuron morphologies at the whole brain level offer exciting opportunities to study single cell genotyping and phenotyping. However, it remains challenging to define cell types and sub-types properly.ResultsAs morphological feature spaces are often too complicated to classify neurons, we introduce a method to detect the optimal subspace of features so that neurons can be well clustered. We have applied this method to one of the largest curated databases of morphological reconstructions that contains more than 9,400 mouse neurons of 19 cell types. Our method is able to detect the distinctive feature subspaces for each cell type. Our approach also outperforms prevailing cell typing approaches in terms of its ability to identify key morphological indicators for each neuron type and separate superclasses of these neuron types. the subclasses of neuronal types could supply information for brain connectivity and modeling, also promote other analysis including feature spaces.AvailabilityAll datasets used in this study are publicly available. All analyses were conducted with python package Scikitlearn 0.23.1 version. Source code used for data processing, analysis and figure generation is available as an open-source Python package, onhttps://github.com/SEU-ALLEN-codebase/ManifoldAnalysisContactljliu@braintell.org
Publisher
Cold Spring Harbor Laboratory