Back to search

Georgia Institute of Technology

Network Optimizations for Literature-based Discovery

Abstract

dc:description.abstract

The main focus of this work pertains to the use of networks, network analysis, and network visualization techniques to improve the speed and efficiency of automated literature-based discovery. This goal is ultimately achieved through the creation of two tools: SemNet (version 2) and CompositeView. SemNet, briefly, is a software tool designed to both create and analyze biomedial knowledge graphs for the purpose of literature-based discovery. This is accomplished through the use of HeteSim, an algorithm designed to evaluate the similarity between nodes and, subsequently, allow them to be ranked based on relevance. In this work, SemNet has been updated in three predominant ways: (1) the knowledge graph data structure has been improved; (2) new algorithms have been developed for improved calculation speed, especially for longer metapaths; (3) ULARA, a flawed rank aggregation system, has been replaced with new aggregation algorithms. SemNet version 2, with the new changes implemented, was used to analyze a use case example where the potential connections between Alzheimer's disease and metabolism were investigated. The same parameters used in this analysis were also applied to SemNet version 1, and the timing results of both analyses were compared. This comparison resulted in SemNet version 2 performing multiple magnitudes better than SemNet version 1. Such improvements make HeteSim calculations using metapaths greater than length two possible with SemNet version 2, thus increasing the potential for more nuanced literature-based discovery. The second tool described in this work, CompositeView, is a visualization software originally designed to visualize SemNet results that has since been generalized to visualize comparable data, network-based or otherwise. These comparable data fall into the ``composite score" definition: large sets of data comprised of smaller sets of data that are conceptually and statistically similar that, when combined, generate a single value with the potential of reducing information overload. The results of many network analysis methods, such as link prediction and relevance analysis (of which SemNet is a part), have the potential to fall into this definition, along with other data not directly associated with networks. Beyond visualization, CompositeView introduces interactive elements through an application framework that allow the user to both filter nodes, edges, and types and adjust the visualization through color changes, node spread, and simulation manipulations. In this work, Compositeview's development process is discussed, along with stress testing and sample data visualizations, all of which help give potential users a good idea of what the tool is capable of. In combination with SemNet, CompositeView provides a capstone to a more consolidated, efficient, and user-friendly literature-based discovery pipeline.

Degree

thesis:*
Level thesis:degree_level
Masters
Department dc:contributor.department
Biomedical Engineering (Joint GT/Emory Department)
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2022

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Allegri, Stephen Anthony
Advisor dc:contributor.advisor
  • Mitchell, Cassie S.
Committee members dc:contributor.committeemember
  • Chau, Duen Horng
  • Fernandez, Todd

Subjects

dc:subject × 7

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/72447

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Allegri, Stephen Anthony. Network Optimizations for Literature-based Discovery. Masters thesis, Georgia Institute of Technology, 2022. https://hdl.handle.net/1853/72447