Multi-scale network analyses for cancer omics data

Abstract

Cancer is a complex disease that results from alterations in genes that control the growth and division mechanisms of the cell. Identifying cancer-associated genes and pathways through which these genes influence cellular function is the key challenge in cancer re- search. This thesis proposes novel computational approaches that integrate data using networks at different scales. Since cancer is a multifactorial disease, integrating multiple data is required to obtain a comprehensive understanding of cancer mechanisms. We introduce a strategy that integrates sample specific data with network-based data. We integrated a protein interaction network with insertional mutagenesis data and discovered that the list of putative cancer genes can be expanded by considering genes that harbor frequent mutations in their interaction network context. As biological mechanisms and genomic events are hierarchically organized, network data should also be analyzed while taking this hierarchy into account. This is called a scale-aware analysis. We propose to apply diffusion kernels on discrete spaces to analyze biological networks in a multi-scale fashion. This process essentially performs a network smoothing. We show that the multi-scale analysis is essential to determine subnetworks of putative cancer genes in their interaction context and that these are most enriched for cancer related pathways. The unfolded human DNA is about 2 meters and is densely packed in the cell nucleus, which influences cell’s functions. We show that the 3D structure of the genome plays an important role whilst identifying genes that are targeted by genomically distal mutations in an integrated insertional mutagenesis screen. We further show that scale- aware topological measures of the chromatin interaction network can predict correlation patterns between genes in the mouse cortex.