Author

Ted Laderas

Date

3-2014

Document Type

Dissertation

Degree Name

Ph.D.

Department

Dept. of Medical Informatics and Clinical Epidemiology

Institution

Oregon Health & Science University

Abstract

HER2-­‐positive breast cancer is an aggressive subtype of breast cancer, with patients having significantly lower survival rates compared to other breast cancers. HER2-­‐positive breast cancer shows pathway addiction to multiple mitogenic signaling pathways. One strategy in cancer treatment is targeted drug therapy, which inhibits the function of specific proteins along these signaling pathways. Predicting drug sensitivity using a patient’s tumor markers such as somatic mutations and copy number variation can help guide treatment selection towards realization of “precision” medicine. The motivation for this work is to (1) characterize the degree of cross-­‐phenotype response when a targeted inhibitor is applied and (2) identify potential oncogene collaborations that can drive the decisions of which targeted treatments should be used. These two problems form the basis of my dissertation aims. In this Dissertation, we highlight two approaches: characterizing drug response and resultant system cross-­‐phenotype response using proteomics and a new network based model of oncogenic collaboration using integration of multiple data types (mutation calls, drug sensitivity, and copy number calls) for visualization... In terms of proteomics, we specifically highlight a robust measure of time series that I apply to proteomics data: Area Under the Curve(AUC). AUC is a measure of upregulation/downregulation over time and I show that it is a useful feature for estimating cross-­‐phenotype response. Specifically, we show that the AUC of four proteins (BIM, RB1, ERK, S6) is highly correlated with drug sensitivity. Using AUCs, I then attempted to predict protein cross-­‐phenotype response for four different cancer cell lines (UACC812, BT549, BT20, and MCF7) using a linear modeling approach called Partial Least Squares Path modeling (PLS-­‐PM). PLS-­‐PM modeling highlights proteins which are well characterized by the data (EGFR, HER2, SRC), but also highlights proteins for which we have incomplete knowledge (PDK1, PTEN, MAPK1) In the network-­‐based approach, we highlight a possible method for integrating unique mutations in a cell line, that of “surrogate mutations”. This method may point the way to precision medicine in a patient specific context. We observe that when imposed on a protein-­‐ protein interaction network, mutations tend to cluster around certain nodes, and show these mutations are statistically significant under a null model of randomly mutated networks. More importantly, we show that these surrogate mutations are associated with drug sensitivity through the use of Random Forests for classification, with an average error rate across all drugs of 30.9%, comparable to the error rate for expression-­‐based subtype specificity of 29.1%. This suggests that surrogate mutations capture network specific elements that are important to predicting drug sensitivity.

Identifier

doi:10.6083/M4J38QVR

School

School of Medicine

Share

COinS
 
 

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.