Development and application of computational methods for cancer subtype detection from -omics data

Abstract

Cancer is a complex and dynamic disease manifesting in ~100 distinct cancer types that arise in multiple cell types and organs due to different but related mechanisms. Research from the last decade has revealed vast heterogeneity within and between cancer types, hampering effective treatment and calling for more personalized treatment strategies. This thesis develops methodology for detection and molecular characterization of cancer subtypes by focusing on the analysis of experiments generating large datasets. The first objective was to provide algorithms for rapid detection and quantification of microRNAs and analysis and visualization of DNA methylation. The second objective was to investigate the cellular and molecular origin of embryonal rhabdomyosarcoma (ERMS), a rare and aggressive childhood cancer.Two new computational methods were implemented and evaluated by comparison to previously published findings. DNA copy number alterations and gene expression estimates were obtained from a novel model system for ERMS and integrated with molecular data from cancer patients. Cell tracing experiments unambiguously demonstrated that ERMS is derived from tissue-resident muscle stem cells, at least in the model system used. In-depth data analysis revealed a diverse molecular basis of ERMS, confirming cancer heterogeneity. Surprisingly, activation of zygotic Dux factors identified a novel cancer subtype that is not limited to ERMS, but occurs in a broad range of different human cancer. Based on the results, it can be concluded that computational methods and integrative data analysis are useful to delineate the origin of cancer subtypes and provide a valuable starting point for selection of relevant therapeutic targets. However, future research is needed to establish more holistic analysis approaches and transfer findings into existing clinical routines.