Knowledge Representation and Data Mining of Neuronal Morphologies Using Neuroinformatics Tools and Formal Ontologies

Abstract

Neuroscience can greatly benefit from using novel methods in computer science and informatics, which enable knowledge discovery in unexpected ways. Currently one of the biggest challenges in Neuroscience is to map the functional circuitry of the brain. The applications of this goal range from understanding structural reorganization of neurons to applying them for smart brain-inspired technology. Mining and comprehending information from micro and/or macro level data generated at various spatial and temporal resolutions is crucial to these goals. This research proposes analytical and search tools that contribute towards a more complete understanding of functional circuitry by transforming complex biological information into useful applied knowledge. As more and more data is being generated informatics tools becomes indispensable. The first aim of my research introduces a Neuroinformatic tool called L-Measure (LM) for quantifying large-scale neuroanatomical data, which is not only particularly applicable to neuronal morphological reconstructions, but also to any other generic tree shaped 3D structural data (e.g., angiography, glial processes). One of the main functions of L-Measure is comparative geometrical and topological analyses of groups of neurons. Community-contributed digitally reconstructed neurons available at NeuroMorpho.Org constitute a “big data” research opportunity for neuroscience discovery beyond the approaches typically pursued in single laboratories. L-Measure is expanded and scaled for a database-wide statistical analysis for investigating robust morphological patterns among heterogeneous neuronal reconstructions. In addition, its development, maintenance and routine usage for wide range of data stands as an exemplar for providing a sustainable software resource to the field. The second aim of this research focuses on enabling knowledge discovery through smart context-based searches as opposed to string-based searches on the shared experimental metadata. Neuroanatomical data that is annotated based on inconsistent terminologies in the literature has limited means for re-use or integration. To solve this problem, a novel approach of representing metadata as machine-readable hierarchies is proposed. The hierarchies represented as formal ontologies constitute the knowledge-base for ontology-based search engine OntoSearch for mining thousands of reconstructions in NeuroMorpho.Org. Applying hierarchical search logic and semantic web technologies, OntoSearch provides direct and fuzzy matches to the primary metadata terminologies. Facilitated by a simple auto-complete enabled search bar, OntoSearch enhances data visibility by at least three times compared to the traditional relational database driven querying.