Huang, Grace T.
(2014)
An Integrated, Module-based Biomarker Discovery Framework.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Identification of biomarkers that contribute to complex human disorders is a principal and challenging task in computational biology. Prognostic biomarkers are useful for risk assessment of disease progression and patient stratification. Since treatment plans often hinge on patient stratification, better disease subtyping has the potential to significantly improve survival for patients. Additionally, a thorough understanding of the roles of biomarkers in cancer pathways facilitates insights into complex disease formation, and provides potential druggable targets in the pathways.
Many statistical methods have been applied toward biomarker discovery, often combining feature selection with classification methods. Traditional approaches are mainly concerned with statistical significance and fail to consider the clinical relevance of the selected biomarkers. Two additional problems impede meaningful biomarker discovery: gene multiplicity (several maximally predictive solutions exist) and instability (inconsistent gene sets from different experiments or cross validation runs).
Motivated by a need for more biologically informed, stable biomarker discovery method, I introduce an integrated module-based biomarker discovery framework for analyzing high- throughput genomic disease data. The proposed framework addresses the aforementioned challenges in three components. First, a recursive spectral clustering algorithm specifically
4
tailored toward high-dimensional, heterogeneous data (ReKS) is developed to partition genes into clusters that are treated as single entities for subsequent analysis. Next, the problems of gene multiplicity and instability are addressed through a group variable selection algorithm (T-ReCS) based on local causal discovery methods. Guided by the tree-like partition created from the clustering algorithm, this algorithm selects gene clusters that are predictive of a clinical outcome. We demonstrate that the group feature selection method facilitate the discovery of biologically relevant genes through their association with a statistically predictive driver. Finally, we elucidate the biological relevance of the biomarkers by leveraging available prior information to identify regulatory relationships between genes and between clusters, and deliver the information in the form of a user-friendly web server, mirConnX.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
9 January 2014 |
| Date Type: |
Publication |
| Defense Date: |
30 September 2013 |
| Approval Date: |
9 January 2014 |
| Submission Date: |
16 December 2013 |
| Access Restriction: |
1 year -- Restrict access to University of Pittsburgh for a period of 1 year. |
| Number of Pages: |
154 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
School of Medicine > Computational Biology |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Biomarker Discovery, Clustering, Spectral Clustering, Variable Selection, Markov Blanket, Group Variable Selection, Regulatory Networks, miRNA |
| Date Deposited: |
09 Jan 2014 15:59 |
| Last Modified: |
15 Nov 2016 14:16 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/20303 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}