Computational approaches for dissecting cancer pathways from insertional mutagenesis data

Computational approaches for dissecting cancer pathways from insertional mutagenesis data

De Ridder, J.

Reinders, M.J.T. (promotor)

Electrical Engineering, Mathematics and Computer Science

Mediamatics

2011-01-31

Advances in the field of molecular biology have resulted in a decent understanding of the causes for and mechanisms through which healthy cells can develop into cancer cells. It is, for instance, well established that cancer is caused by mutation of so-called cancer genes. That said, current knowledge on exactly which genes can function as a cancer gene is far from complete. Fur- thermore, the regulatory pathways through which cancer genes exhibit their malicious effect on healthy cell division remain largely elusive. To identify novel cancer genes, insertional mutagenesis screens can be employed. In these screens tumors are induced by viral mutations in the DNA of a mouse. Since these mutations are likely to be observed in the vicinity of cancer genes this is a fruitful method of cancer gene discovery. The computational approaches proposed in this thesis are primarily aimed at analyzing in- sertional mutagenesis data. To detect commonly mutated regions in the mouse genome, which point to novel cancer genes, we employ a kernel convolution framework. The major advan- tage of this method is that the data is analyzed in a scale-space, which allows the detection of regions of various widths. Furthermore, the probability of making an error is controlled inde- pendent of the width of a commonly mutated region, making the framework suitable for the analysis of large screens. This framework can also be applied to detect commonly co-occurring mutations, which reveal possible collaboration between cancer genes. We also elaborate on additional applications and generalizations of the scale-space framework for analyzing other types of biomolecular data. To delineate the pathways through which cancer genes act, we propose a mutational ge- nomics approach. To this end, the mutation data is complemented with gene expression data measured in the same samples. This enables the inference of associations between the presence or absence of an insertion and the gene expression. In this thesis we explore the use of Boolean association models that combine multiple mutated loci to predict gene expression levels. These models are also applied to a genetical genomics dataset. The discovered associations provide insight into how (cancer) genes are connected in cellular regulatory pathways.

bioinformatics

http://resolver.tudelft.nl/uuid:b7cc3811-2413-42d7-b33f-d20209bb040d

Mediamatica

9789490818067

Institutional Repository

doctoral thesis

(c) 2011 De Ridder, J.