Computational methods for personalized cancer genomics

Hou, Jack Pu

Permalink

https://hdl.handle.net/2142/97521 Copy

Description

Title

Computational methods for personalized cancer genomics

Author(s)

Hou, Jack Pu

Issue Date

2017-04-21

Director of Research (if dissertation) or Advisor (if thesis)

Ma, Jian

Doctoral Committee Chair(s)

Milenkovic, Olgica

Committee Member(s)

• Song, Jun
• Hergenrother, Paul

Department of Study

Bioengineering

Discipline

Bioengineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Personalized medicine
• Cancer
• Genomics
• Drug
• Prescription
• Machine learning
• Gene expression
• Mutation
• Copy number
• Breast cancer
• Correlation
• Clustering
• Gene pathway
• Drug target

Abstract

In recent years, cancer treatment strategies have moved towards personalized approaches, specifically tailoring cancer treatments on a single-patient basis using molecular profiles from the patients’ tumor genomes. Knowledge of a patient’s molecular profile can be used to 1) identify the disease mechanisms and underlying cause of a single patient’s cancer, 2) assign patients into treatment groups based on the molecular prognosis, and 3) recommend potential treatments for individual patients based on the patient’s molecular signature data. However, the bottleneck of the personalized medicine approach lies in the challenge of translating the vast amount of sequencing data to meaningful clinical insights. This dissertation explores several computational methods that utilize molecular signature data to understand disease mechanisms of cancer, categorize patients into biologically relevant subtypes, and recommend drug treatments to patients. In the dissertation, we present a method, DawnRank, a patient-specific method that determines the potential driving genomic alterations (the drivers) of cancer. We expand on DawnRank’s capabilities by using the DawnRank scores in key driver mutations and copy number variants (CNVs) to identify breast cancer subtypes. We found 5 alternative subtypes based on potentially clinically relevant driver genes, each with unique defining target features and pathways. These subtypes correspond to and build upon our previous knowledge of breast cancer subtypes. We also identify disease mechanisms in identifying key novel cancer pathways in which driver genes interact. We developed a method, C3, which pinpoints patterns of cancer mutations in a pathway context from a patient population to detect novel cancer pathways that consist of significant driver genes. C3 improves on current methods in driver pathway detection both on a technical aspect and a results-oriented aspect. C3 can detect larger and more consistent pathways than previous methods as well as discovering more biologically relevant drivers. Finally, we address the issue of drug recommendation in the wake of molecular signature data. We develop a method, Scattershot, which combines genomic information along with biological insights on cancer disease mechanisms to predict drug response and prioritize drug treatments. Scattershot outperforms previous methods in predicting drug response and produces recommendations that largely comply with known medical treatment protocols.Scattershot recommends drugs to cancer patients that are in line with the actual drugs prescribed by the physician.

Graduation Semester

2017-05

Type of Resource

text

Permalink

http://hdl.handle.net/2142/97521

Copyright and License Information

Copyright 2017 Jack Hou

Owning Collections