Author ORCID Identifier


Date of Graduation


Document Type

Dissertation (PhD)

Program Affiliation

Biomedical Sciences

Degree Name

Doctor of Philosophy (PhD)

Advisor/Committee Chair

Shuxing Zhang

Committee Member

Alemayehu Gorfe

Committee Member

Ju-Seog Lee

Committee Member

Peng Wei

Committee Member

Zhongming Zhao

Committee Member

Jordi Rodón


Three of the major problems facing cancer therapeutics are 1) drug resistance, the intrinsic or acquired ability of cancer cells to evade the effect of the therapies used to treat them; 2) heterogeneity among individual patients’ disease at the molecular level and the resulting variability in therapeutic response; and 3) the limitations of genomics biomarkers in matching patients to the most effective therapy. One possible solution to drug resistance is the use of combination therapies rather than monotherapies. Use of multiple drugs, each with a different mechanism of action, lowers the chances that the cancer cells will develop or have intrinsic resistance to all the drugs in a given treatment regimen. However, identifying effective drug combinations is a daunting task, since there are over 2000 FDA-approved drugs alone, making the physical testing of even all possible pairwise combinations unfeasible, let alone higher-order drug combinations or combinations with investigational drugs. Furthermore, as mentioned, different patients will respond differently to the same therapy, and it is becoming apparent that effective treatment will require a personalized-medicine approach, or treating patients according to the individual molecular characteristics of their respective diseases, adding another layer of complexity to the problem. Advances in computing technology have made this a challenge amenable to computational approaches, and in this work, we develop a computational framework for predicting the most effective drug combinations for an individual patient using the gene-expression profile of the patient’s disease. The use of gene-expression profiles rather than genomics biomarkers in therapy selection allows a wider range of patients to benefit, rather than only the patients with actionable biomarkers.

We begin by developing a pipeline for processing individual gene-expression data from any platform and extracting information about the transcriptional dysregulation of not only individual genes but also key pathways that may be driving the disease biology. Next, we develop a system for incorporating drug-target and pathway data from several different sources and predicting the perturbational effects of drugs on transcription. We then use these processed data—patient gene-expression and drug-perturbational—as inputs for a machine-learning model to predict the in vitro effects of drug combinations on cell lines or patient samples. Finally, we expand our scope from pathways to entire biological networks and develop a knowledge-based scoring system to predict drug-combination efficacy in individual patients based on network biology. Taken together, this work presents a biologically informed, efficacious computational framework for identifying effective drug combinations for patients based on their gene-expression profiles.


precision medicine, personalized medicine, drug combinations, computational, gene expression, network biology, machine learning



To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.