Date of Award

Spring 2015

Document Type


Degree Name

Doctor of Philosophy in Computing Sciences - (Ph.D.)


Computer Science

First Advisor

Yehoshua Perl

Second Advisor

James Geller

Third Advisor

Michael Halper

Fourth Advisor

Huanying Gu

Fifth Advisor

Gai Elhanan

Sixth Advisor

James J. Cimino

Seventh Advisor

M. Rusinkiewicz


Biomedical ontologies are complex knowledge representation systems. Biomedical ontologies support interdisciplinary research, interoperability of medical systems, and Electronic Healthcare Record (EHR) encoding. Ontologies represent knowledge using concepts (entities) linked by relationships. Ontologies may contain hundreds of thousands of concepts and millions of relationships. For users, the size and complexity of ontologies make it difficult to comprehend “the big picture” of an ontology's content. For ontology editors, size and complexity make it difficult to uncover errors and inconsistencies. Errors in an ontology will ultimately affect applications that utilize the ontology.

In prior studies abstraction networks (AbNs) were developed to provide a compact summary of an ontology's content and structure. AbNs have been shown to successfully support ontology summarization and quality assurance (QA), e.g., for SNOMED CT and NCIt. Despite the success of these previous studies, several major, unaddressed issues affect the applicability and usability of AbNs. This thesis is broken into five major parts, each addressing one issue.

The first part of this dissertation addresses the scalability of AbN-based QA techniques to large SNOMED CT hierarchies. Previous studies focused on relatively small hierarchies. The QA techniques developed for these small hierarchies do not scale to large hierarchies, e.g., Procedure and Clinical finding. A new type of AbN, called a subtaxonomy, is introduced to address this problem. Subtaxonomies summarize a subset of an ontology's content. Several types of subtaxonomies and subtaxonomy-based QA studies are discussed.

The second part of this dissertation addresses the need for summarization and QA methods for the twelve SNOMED CT hierarchies with no lateral relationships. Previously developed SNOMED CT AbN derivation methodologies, which require lateral relationships, cannot be applied to these hierarchies. The Tribal Abstraction Network (TAN) is a new type of AbN derived using only hierarchical relationships. A TAN-based QA methodology is introduced and the results of a QA review of the Observable entity hierarchy are reported.

The third part focuses on the development of generic AbN derivation methods that are applicable to groups of structurally similar ontologies, e.g., those developed in the Web Ontology Language (OWL) format. Previously, AbN derivation techniques were applicable to only a single ontology at a time. AbNs that are applicable to many OWL ontologies are introduced, a preliminary study on OWL AbN granularity is reported on, and the results of several QA studies are presented.

The fourth part describes Diff Abstraction Networks, which summarize and visualize the structural differences between two ontology releases. Diff Area Taxonomy and Diff Partial-area Taxonomy derivation methodologies are introduced and Diff Partial-area taxonomies are derived for three OWL ontologies. The Diff Abstraction Network approach is compared to the traditional ontology diff approach.

Lastly, tools for deriving and visualizing AbNs are described. The Biomedical Layout Utility Framework is introduced to support the automatic creation, visualization, and exploration of abstraction networks for SNOMED CT and OWL ontologies.