Discovering biomarkers of drug efficacy in cancer from pharmacogenomic data
Kalamara, Angeliki; Blank, Lars M. (Thesis advisor); Saez-Rodriguez, Julio (Thesis advisor)
Dissertation / PhD Thesis
Dissertation, RWTH Aachen University, 2020
Cancer is a biologically complex disease with clinically diverse outcomes. Successful therapies are most often hampered by the observed high molecular heterogeneity of this disease. For these reasons, effective cancer treatment is still a challenge. Nowadays, it is clear that a cancer therapy that fits all cases cannot be found, and as a result there is a pressing need of methods to tailor therapeutic strategies on a single patient level, based on the molecular features of their cancers. Pharmacogenomics aims to study the relationship between an individual's genotype and drug response. Scientists use different biological models, ranging from cell lines to mouse models, as proxies for patients for preclinical and translational studies. The rapid development of "-omics" technologies is increasing the amount of features that can be measured in these models, expanding the possibilities of finding predictive biomarkers of drug response. Uncovering these relationships requires diverse computational approaches ranging from machine learning to dynamic modeling. Despite major advances, we are still far from being able to precisely predict drug efficacy in cancer models, let alone directly on patients. Here, I deal with the topic of in vivo drug response prediction and the discovery of predictive biomarkers of drug response approached in two different ways. Firstly, I integrate publicly available gene expression profiles of immortalised human cancer cell lines and primary tumor samples in order to bridge rich pharmacogenomic data derived from the former to the latter, thus predicting drug sensitivity in vivo. I apply this pipeline in the context of breast and colon cancer and validate the predictions across independent datasets. This study provides a conceptual framework to tackle the problem of predicting anti-cancer drug response in vivo via public data integration and defines a general strategy applicable to multiple cancer types. Subsequently, I shift from the aforementioned data-driven analysis to a network-based approach, in an attempt to add a more mechanistic insight to the link between a biomarker and the corresponding drug. To that end, I use the tool CARNIVAL which is based on causal reasoning principles. Basal gene expression from cancer cell lines were combined with a prior knowledge network, transcription factor and pathway activities in an attempt to reconstruct pathways linked with drug sensitivity. Despite the challenges, my analysis based on cancer-specific gene expression analysis could yield predictive biomarkers of drug response that could be validated consistently across different independent datasets in both cancer types. I envision that the framework presented here can contribute to generate new biomarker-drug hypotheses in various types of cancer, aid pharmacogenomic discovery and elucidate their clinical implications.