The study of cancer represents one of the most important areas of research in the field of human diseases. In spite of the enormous resources invested in biomedical research in recent years, it still represents one of the pathologies with the highest mortality rate. Cancer describes a group of diseases whose common factor lies in the uncontrolled growth of a group of cells known as tumours. These pathologies have a very heterogeneous genetic basis, even when comparing individuals with the same type of cancer. This makes the search for common markers that can later be used in the design of general-purpose therapies a challenge. In this context, computational biology (CB) is one of the most valuable tools, providing an enormous variety of statistical methods for the quantification and analysis of molecular pathways, which describe how protein-protein interactions occur to carry out essential functions in the cell. This approach has enabled the study of cancer to be addressed through systems biology, a discipline that describes diseases as specific alterations in parts of a system, beyond the changes produced in its individual elements. Notably, this approach has allowed us to understand why alterations in different proteins cause similar changes in molecular pathways, of great relevance in the study of genomic variability in cancer. The aim of this work is the study of genomic heterogeneity in cancer. To this end, a general analysis protocol is described, starting with the estimation of somatic mutations, followed by an assessment of their effect on the different signalling pathways. In this case, a hierarchical factorisation model will be used to simultaneously determine the latent components, both at the gene level and at the molecular pathway level. These components will correspond to the different survival strategies implemented in real tumours, which will provide a global view of genomic heterogeneity from different abstraction levels, including mutations, genes, pathways and, finally, the biological characteristics shared by all cancers, known as cancer hallmarks. Chapter 1 begins with a general introduction to the context of this thesis. First, a general description of the hallmarks of cancer will be given, which will have an impact on the design of the statistical models that will be proposed in the following chapters. Chapter 2 describes the implementation of a computational protocol that aims at the robust prediction of somatic mutations in a group of cancer patients. For this purpose, the modelling of noise and its application to estimate the expected degree of error in each genomic region will be of particular relevance. Chapter 3 aims to study the genomic heterogeneity observed between patients from a systems biology perspective. To this end, a hierarchical factorisation model is described allowing the simultaneous derivation of a set of latent components at the gene level and their corresponding components at the molecular pathway level. Finally, chapter 4 offers a series of general conclusions of this thesis, emphasising the advantages of a protocol of analysis such as the one described in this thesis and its application in real cases. In addition, some future lines of work will be described that will allow further development of the different methodologies proposed in this work.
