UVM Theses and Dissertations
Format:
Print
Author:
Nabhan, Ahmed Ragab
Dept./Program:
Computer Science
Year:
2014
Degree:
PhD
Abstract:
Recent advances in high throughput technologies have led to an increasing amount of rich and diverse biological data and related literature. Model organisms are classically selected as subjects for studying human disease based on their genotypic and phenotypic features. A significant problem with model organism identification is the determination of characteristic features related to biological processes that can provide insights into the mechanisms underlying diseases. These insights could have a positive impact on the diagnosis and management of diseases and the development of therapeutic drugs. The increased availability of biological data presents an opportunity to develop data mining methods that can address these challenges and help scientists formulate and test data-driven hypotheses.
In this dissertation, data mining methods were developed to provide a quantitative approach for the identification of potential model organisms based on underlying features that may be correlated with disease manifestation in humans. The work encompassed three major types of contributions that aimed to address challenges related to inferring information from biological data available from a range of sources. First, new statistical models and algorithms for graph pattern mining were developed and tested on diverse genres of data (biological networks, drug chemical compounds, and text documents).
Second, data mining techniques were developed and shown to identify characteristic disease patterns (disease fingerprints), predict potentially new genetic pathways, and facilitate the assessment of organisms as potential disease models. Third, a methodology was developed that combined the application of graph-based models with information derived from natural language processing methods to identify statistically significant patterns in biomedical text. Together, the approaches developed for this dissertation show promise for summarizing the information about biological processes and phenomena associated with organisms broadly and for the potential assessment of their suitability to study human diseases.
In this dissertation, data mining methods were developed to provide a quantitative approach for the identification of potential model organisms based on underlying features that may be correlated with disease manifestation in humans. The work encompassed three major types of contributions that aimed to address challenges related to inferring information from biological data available from a range of sources. First, new statistical models and algorithms for graph pattern mining were developed and tested on diverse genres of data (biological networks, drug chemical compounds, and text documents).
Second, data mining techniques were developed and shown to identify characteristic disease patterns (disease fingerprints), predict potentially new genetic pathways, and facilitate the assessment of organisms as potential disease models. Third, a methodology was developed that combined the application of graph-based models with information derived from natural language processing methods to identify statistically significant patterns in biomedical text. Together, the approaches developed for this dissertation show promise for summarizing the information about biological processes and phenomena associated with organisms broadly and for the potential assessment of their suitability to study human diseases.