Computational and Experimental Approaches to Cancer Biomarker Discovery

Effective cancer treatment requires good biomarkers: measurable indicators of some biological state or condition that constitute the cornerstone of personalized medicine. Prognostic biomarkers provide information about the likely course of the disease, while predictive biomarkers enable prediction of a patient’s response to a particular treatment, thus helping to avoid unnecessary treatment and unwanted side effects in non-responding individuals.Currently biomarker discovery is facilitated by recent advances in high-throughput technologies when association between a given biological phenotype and the state or level of a large number of molecular entities is investigated. Such associative analysis could be confounded by several factors, leading to false discoveries. For example, it is assumed that with the exception of the true biomarkers most molecular entities such as gene expression levels show random distribution in a given cohort. However, gene expression levels may also be affected by technical bias when the actual measurement technology or sample handling may introduce a systematic error. If the distribution of systematic errors correlates with the biological phenotype then the risk of producing false positive biomarkers increases. Therefore, understanding the sources of bias and removing it is essential for effective biomarker discovery.The first part of this thesis describes a tool for visualization of technical bias in the microarray data. The researcher can readily see whether a dataset of interest is biased, and whether the bias correction method used is effective to correct it. Thus the potential value of the various microarray data sets can be evaluated.Mutational signatures constitute a particularly attractive and robust class of biomarkers. They characterize and quantify at least two fundamental mechanisms responsible for DNA aberrations present in a given tumor: 1) active mutational processes caused either by endogenous or exogenous factors, for example chemical agents such as tobacco smoke or cancer cytotoxics, or by active enzymatic processes such as APOBEC related mutagenesis; and 2) the integrity of endogenous DNA repair processes as exemplified by BRCA1/2 dysfunction or MMR deficiency. Since lack of a given DNA repair process may make tumors particularly sensitive to certain types of therapy, identification of such defects will allow for potential enhancements of the therapy efficacy. State of the art mutational signatures are derived mathematically using nonnegative matrix factorization to solve a blind source separation problem arising from a multitude of mutational processes that form the observable mutational catalogs. In my ongoing projects I address this issue with a purely biological, experimental approach where the effects of treatment with cytotoxic agents or defects in DNA repair mechanisms can be individually quantified and turned into mutational signatures.In the second part of the thesis I present work towards identification and improvement of the current mutational signatures through an experimental approach. For that purpose a unique  chicken cell line, DT40, was chosen, that allows for a relatively easy, HR-mediated knockout of DNA repair genes. In order to effectively use the DT40 in the subsequent mutational signatures analysis the DT40 genome was sequenced, assembled and characterized, which is described in the thesis. We are currently using it as a model system in our framework for functional analysis study of DNA repair mechanisms and cytotoxic effects. We hope that the experimentally derived mutational signatures will be useful as a part of patient diagnostics in the future. It is here that we had to focus our attention to various sources of biological bias while trying to address a particular demand within cancer therapy. This part of the thesis describes our ongoing efforts; thus by the time of the defense some updates to this part are expected.This work, together with manifold of efforts being done all over the world, is hopefully a step towards implementation of personalized medicine and better treatments for cancer patients.