Neuroscience is a discipline rich in data. In neuroscience, information about the nervous system is derived using a wide-range of experimental methodologies. While each of these methodologies can provide unique insight into nervous system function, each of them are also fundamentally tied to the subjects they measure. Because of this, no single experimental methodology can provide a universal perspective on nervous system function. It becomes sensible, therefore, to explore how we might gain greater insight into the nervous system from unifying the data derived from different experimental methodologies.

In this dissertation, I apply techniques from computer science to assemble the rich sets of data in neuroscience into data structures that allow otherwise impractical or impossible questions to be answered. Because of the nature of the nervous system, these data structures must be capable of dealing with data describing parts of the nervous system that vary in size across eight orders of magnitude, otherwise known as "multi-scale" data. These parts include molecules, parts of cells, cells and their circuits, brain regions and features of gross anatomy.

In order to address this challenge, I distinguish between information about the nervous system that expresses facts or semantic relationships (the "what"), from information that expresses images or spatial information (the "where"). I apply a data structure known as an ontology to describe parts of cells relevant to neuroscience, i.e. subcellular anatomy. This ontology is applied to enable automated question answering, i.e. inference, from data sets displaying parts of cells that are derived from electron microscopy. The challenges of building these data structures as a neuroscience community are reviewed, and in response I present an online software system, NeuroLex.org, that provides a first step towards addressing them. Additional examples of automated question answering in neuroanatomy are shown using the NeuroLex knowledge base, which now spans ~17,000 concepts specific to neuroscience.

To address the challenge of the "where" of neuroscience, I present an online software system, the Whole Brain Catalog, that enables two-dimensional, three-dimensional, and four-dimensional (i.e. time-varying) multi-scale data sets to be spatially integrated within a common coordinate system. These data are then navigable by a user in three dimensions in real time and across spatial scales, and can be integrated to answer scientific questions. The dissertation concludes with a look towards the future of the continued convergence between computer science and neuroscience.

This dissertation presents the results of three efforts to build practical systems for formalizing concepts in science:neuron types, brain regions, and experimental protocols. Neuron types and brain regions are foundational concepts in neuroscience, and protocols are foundational for all scientific results and the concepts we build from them.

Chapter One presents the Neuron Phenotype Ontology and supporting tools and their application to model common types generallyknown in the field and to experimental types defined by the exact techniques employed in a single lab. The goal: to provide a common method to name and communicate about neuron types by composing types as collections of phenotypes with each phenotype being a pair of a term for the value of the phenotype drawn from existing shared terminology and a relationship that captures the data modality of and methodology used to determine that value.

Chapter Two presents the AtOM ontology model for anatomical atlases and the results of applying it to model a wide rangeof extant atlases. The goal: to establish standard ways of identifying atlases and their versions and enabling their use in digital infrastructure to facilitate a wide variety of use cases. Examples include clear communication about the exact spatial and semantic brain regions from which experimental data are collected and linking those regions to the methodologically defined criteria used to delineate their boundaries.

Chapter Three presents protc/ur, a domain specific language for specifying protocols, and presents the results of applying it toextract structured data from experimental protocols. The goals: to validate the protc/ur domain model and curation workflows, show that protc/ur enables queries over complex relationships to find quantitative data extracted from natural language, and, ultimately, demonstrate that protc/ur and the system as a whole are an effective way to formalize protocols and make the details of methodology visible in information systems.

Taken together these chapters show the effectiveness of using experimental methodology as an organizing principle in scientificinformation systems and the potential that it has as a guiding principle for building practical tools for working scientists.