Machine-Readable Description of Neuron Types and Properties

David J. Hamilton (George Mason University), Maurizio Bergamino (George Mason University), Javier DeFelipe (Tecnología Biomedica & Instituto Cajal), Nicolas Le Novère (EMBL-European Bioinformatics Institute), Gordon M. Shepherd (Yale Medical School), Menno P. Witter (Norwegian University of Science and Tech), Giorgio A. Ascoli (George Mason University)

Neuron types are in general operationally described based on their properties, e.g. electrophysiological, morphological, molecular, developmental, functional, and connectivity. For example, a kind of cortical interneuron is commonly characterized as fast-spiking, extending a basket-shaped axon, and parvalbumin-containing. Such descriptions are usually expressed in natural language, optimizing intuitive understanding and human communication. For computational consumption, this knowledge must be expressed into machine-readable form. The conversion should be sufficiently flexible to enable wide applicability, balancing rigorous logic, algorithm efficiency, and accessibility to neuroscience experts. To this aim, we propose to define neuron types as collections of properties formulated as part-relation-value (such as soma-has_molecule-parvalbumin). Each property ascribed to a neuron type is necessary, i.e. if a neuron is shown not to possess that property, it does not belong to that class. The collection of properties defining a neuron type is altogether sufficient: a neuron meeting all those conditions is certainly of that type. The terms used as property descriptors must themselves be progressively linked to machine-readable definitions. These principles have been recently adopted in creating the Neuron Registry, an initiative of the International Neuroinformatics Coordinating Facility ( under the Program on Ontologies for Neural Structures (PONS). The goal is an open access resource for comparing potentially new neuronal types with known types and for constructing statistical representations of neuronal cell types based on known instances. A task force of volunteer experts entered neuronal properties for an initial set of neuron types with a cloud-implemented curator interface suitable for crowd-sourcing ( We are now launching the ‘adopt-a-neuron’ campaign: an open invitation to the entire neuroscience community for (self-)nominations of neuron type curators ([email protected]). Submissions of neuron types are encouraged from any species and brain region. Every property must be accompanied by a relevant literature citation and will be peer-reviewed.

Preferred presentation format: Demo
Topic: General neuroinformatics

Document Actions