Cancer is an umbrella terminology that binds hundreds of complex genetic diseases based on a set of common phenotypic hallmarks. Each cancer and their sub-types have their unique genomic profiles. The common factor that binds them all together is that they all arise from changes in the DNA. Theses changes range from single nucleotide levels variation to large scale chromosomal aberrations. The consequences of these changes also have distinct impacts on disease development and progression depending on their ability to alter the protein function. Changes in the DNA of a protein-coding gene might have a directly quantifiable impact while quantifying the impact of a change in the regulatory DNA (viz. noncoding) element is a non-trivial task. A better understanding of the complex interplay between coding and noncoding genetic variation will lead to a better understanding of the diseases and improve diagnostics and patient care. This thesis proposes a novel framework for reliable prediction of somatic point mutations in cancer genomes. The framework was applied to several whole-genome and exome sequenced cancer datasets. Our findings suggested that a consensus-based approach produces a more reliable result than individual mutation detection tools. We also proposed an in-silico post-processing workflow and in-vitro validation guideline to improve the detection accuracy of using orthogonal techniques. Different cancers have distinct mutational burden and profile and understanding these genomic sub-types will lead to better patient stratification and clinical management. Using mutational signature analysis we investigated the inter- as well as intra-tumour heterogeneity in colon adenomas and skin adnexal tumours. By comparing the mutational signature as well as mutation burden between adult and paediatric patients, we identified striking genomic similarities between them. Based on these findings, we recommend that like many adult patients, genomic profiles of paediatric patients should also be routinely taken into consideration while deciding the therapeutic course. Mutations that give selective survival advantages to cancer cells are commonly referred to as driver mutations. These mutations can occur both in the protein-coding region of the genome or beyond it. This thesis reviewed several available driver mutation detection tools and identified a few areas with a considerable scope of improvement. We proposed a novel machine learning-based framework to prioritize noncoding driver mutations in cancer genomes.@en