Morphological Classification of Glia: a Neuroinformatics Approach



Journal Title

Journal ISSN

Volume Title



Neurons and glia are the two main types of cells in the nervous system. Neurons communicate by transmitting signals giving different species the ability to perform various complex functions. As opposed to neurons, glial cells support the nervous system and are responsible for maintaining normal homeostasis. Glial cells are as abundant as neurons in the nervous systems of most animals, including humans. Just like neurons, glial cells are also characterized by complex branching morphologies. Scientists have long been interested in classifying neurons; however, little attention has been given to classifying glia.In recent years, advances in neuroscience research have led to an increased interest in neuroinformatics, data sharing, and online data repositories, as well as a related need for data organization. NeuroMorpho.Org is the world’s largest public repository of digitally reconstructed neural morphology with more than 172,000 traced cells. This online neuralbank provides open access to neural reconstructions so that published data can be downloaded for statistical and mathematical modeling, leading to new scientific discoveries. Each tracing entry in NeuroMorpho.Org is accompanied by a battery of morphometric features such as length, volume, angles, diameter, etc., as well as detailed metadata annotations describing the animal subject, anatomy, and experimental preparation. As a member of the NeuroMorpho.Org team, I have mastered different tasks including database maintenance, and helped NeuroMorpho.Org grow from 62,000 to more than 172,000 digital neural tracings, of which glia now constitute more than 10% of the total content. However, content expansion and continuous growth increased the complexity of data and metadata, requiring effective resources for information access. Therefore, we introduced the online tool Summary reporting to generate structured reports of morphometry organized into homogenous metadata groups for arbitrary subsets of data selected by the user. As the first application of this new functionality, we focused on advancing models to classify neurons and glia. We applied supervised learning algorithms including Support Vector Machine, Random Forest, and K-Nearest Neighbors to distinguish neurons and glia based on their extracted morphometric features. Across a diverse set of metadata, including species, brain regions, and histological processing, the classifiers were able to discern neurons and glia with high precision. Our results indicate that arbor morphology is an effective and robust way to categorize brain cells. In particular, our work identifies an individual morphometric measure, Average Branch Euclidean Length (ABEL), which can be robustly used to distinguish neurons from glia across different vertebrate and invertebrate animal models, a variety of experimental conditions, and anatomical regions, except for the cerebellum. In addition, we found that calculating the ABEL from as few as five branches can provide above 95% accuracy in classifying neurons and glia without requiring the entire cell arbor to be reconstructed.



Classification, Glia, Machine learning, Morphology, Neuroinformatics, NeuroMorpho.Org