Species, brain regions, and cell types: hierarchies for mining neuromorphometry

Organizing, standardizing, and integrating metadata are often necessary steps between data sharing and data mining. This process is particularly challenging in neuroscience, because data comes in a multitude of forms and scales, from structural to functional and from molecules to whole brains. Continuous investigations and interest in neuronal morphology have recently led to densely populated databases and a growing user community seeking intuitive and effective access to digital data. Neurons are typically identified by their animal species, brain regions, and cell types. Organizing the representation of these three primary axes into ontological hierarchies should enable smarter and more powerful search and browse capabilities. As a proof of concept, we developed OntoSearch, a practical approach harnessing standardized terminologies and emerging ontologies for mining neuromorphological data from NeuroMorpho.Org. NeuroMorpho.Org is the largest existing repository of public digital 3D reconstructions of neuronal morphology. It contains over 6000 neurons of 150 unique cell types, contributed by more than 50 laboratories worldwide from 30 different animal strains and 70 brain regions. A standard keyword search, such as ‘C57BL6/129SvJ,’ ‘Macaca fascicularis,’ and ‘F344,’ returns hits because these strings are hard-coded in the database. Additionally, OntoSearch will also recognize queries for ‘SvJ,’ ‘cercopithecinae,’ ‘transgenic mice,’ and ‘rodents,’ by cross-mapping these terms with phylogenetically classified species taxonomy. Unfortunately, transforming the knowledge on anatomical parcellations even for a single animal order (such as rodent or primate) into formal ontologies is more challenging, because, unlike species, no single hierarchical organization (e.g. morphological, developmental, histological, functional, etc.) is accepted or dominant for brain regions. Ontosearch thus determines the correspondence between NeuroMorpho.Org content and multiple available hierarchical partonomies via parallel mapping. This approach also implies the identification of semantically equivalent classes, such as ‘somatosensory deep layers’ and ‘somatosensory layers 5 and 6,’ based on the relationship among immediate parent and child nodes, synonyms, and definitions (ontology alignment). Suitable resources for such comparisons include database and/or XML schemas, taxonomies, dictionaries, and ontology description languages (e.g. OWL and OBO). Aside from rodents and primates, the structures and cell types of the adult fly brain may also be delineated using part_of and has_synaptic_terminals_of relations (virtualflybrain.org). This allows users to find ‘glomerulus DL2’ cells when performing a search for ‘antennal lobe glomerulus.’ Nevertheless, the formal classification of neuron types based on their properties is even more complex than that of brain regions (see poster ‘Machine-Readable Description of Neuron Types and Properties’ by D. Hamilton et al.). Developing Ontosearch within the realm of a particular knowledge domain and a single database enables controlled testing of available semantic hierarchies. At the same time, Ontosearch pragmatically enhances the functionality of NeuroMorpho.Org and provides a useful service to the neuromorphological community. Moreover, this approach may be generally applicable to other neuroinformatic repositories and subfields of neuroscience. In this perspective, the ontological mapping generated in this effort may be valuable to other ongoing attempts to create neuroscience specific formal ontologies.