Mathematical models for data mining and system dynamics to study head and neck cancer progression and chemoprevention

Abstract

Head and neck squamous cell carcinoma (HNSCC) is the 6th most prevalent cancer worldwide, and more than 12,000 deaths from this disease are anticipated in 2015 in the U.S. alone. The advent of the “Big Data” era for biomedicine, through the widespread use of genomic, transcriptomic, and other –omic data acquisition technologies, has enabled deeper exploration of the molecular-level mechanisms behind HNSCC development and progression. This knowledge in turn can lead to earlier diagnosis and better treatment strategies, resulting overall in better patient outcomes. However, the volume and complexity of –omic data present a major obstacle to fully realizing its potential to accelerate and enable basic and translational research for HNSCC. The goal of this Ph.D. dissertation is to address several key technical challenges related to harnessing –omic data for clinical HNSCC research. These are (1) the lack of knowledge-driven modeling tools and systems for discovering biomarkers at the protein and metabolite levels; (2) the lack of effective strategies for integrating heterogeneous types of –omic data for prediction; and (3) the lack of systems-level representations of biomarker knowledge for effectively predicting responses to bioactive agents. This dissertation addresses these challenges through three specific aims: 1. Knowledge-driven Data Mining: To develop modeling tools to mine –omic datasets in HNSCC for biomarker discovery by harnessing existing knowledge 2. Integrated –Omic Modeling: To develop supervised learning models for predicting HNSCC progression through integration of –omic datasets 3. System Modeling: To develop dynamic system models for predicting response to combinations of multi-target agents against HNSCC. The research in this dissertation was completed in collaboration with the Winship Cancer Institute and Georgia Institute of Technology. The models and tools developed have been systematically evaluated and validated using a variety of –omic data types. These results and associated case studies demonstrate the contribution of this work and its future potential in computational HNSCC research.