Biomedical data integration in cancer genomics
Year: 2015
Type: dissertation
Abstract: Cancer is one of the leading causes of death in industrialized nations and its incidence is steadily increasing due to population aging. Cancer constitutes a group of diseases characterized by unwanted cellular growth which results from random genomic alterations and environmental exposure. Diverse genomic and epigenomic alterations separately and jointly regulate gene expression and stimulate and support neoplastic growth. More effective treatment, earlier and more accurate diagnosis, and improved management of cancer are important for public health and well-being. Technological improvements in data measurement, storing and transport capability are transforming cancer research to a data-intensive field. The large increases in the quality and quantity of data for the analysis and interpretation of experiments has made employing computational and statistical tools necessary. Data integration - the combination of different types of measurement data - is a valuable computational tool for cancer research because data integration improves the interpretability of data-driven analytics and can thereby provide novel prognostic markers and drug targets. I have developed two computational data integration tools for large-scale genomic data and a simulator framework for testing a specific type of data integration algorithm. The first computational method, CNAmet, enhances the interpretation of genomic analysis results by integrating three data levels: gene expression, copy-number alteration, and DNA methylation. The second computational method, GOPredict, uses a knowledge discovery approach to prioritize drugs for patient cohorts thereby stratifying patients into potentitally drug-sensitive subgroups. Using the simulator framework, we are able to compare the performance of integration algorithms which integrate gene copy-number data with gene expression data to find putative cancer genes. Our experimental results indicate in simulated, cell line, and primary tumor data that well-performing integration algorithms for gene copy-number and expression data use and process genomic data appropriately. Applying these methods to diffuse large B-cell lymphoma, integrative analysis of copy-number and expression data helps to uncover a gene with putative prognostic utility. Furthermore, analysis of glioblastoma brain cancer data with CNAmet suggests that a number of known cancer genes, including the epidermal growth factor receptor, are highly expressed due to co-occuring alterations in their promoter DNA methylation and copy-number. Finally, integration of publicly available molecular and literature data with GOPredict suggests that treating patients with FGFR inhibitors in breast cancer and CDK inhibitors in ovarian cancer could support standard drug therapies. Collectively, the methods developed here and their application to varied molecular cancer data sets illustrates the benefits of data integration in cancer genomics.
Source: Työväentutkimus Vuosikirja
Author: Riku Louhimo
Language: English
Subfield: Cancer Research
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W1493356594