Mutations are the source of genetic variation in natural populations, provide material for molecular evolution and, importantly, cause human genetic diseases. Yet the mechanisms of mutagenesis are, to date, not completely understood. Low mutation frequencies limit direct observations in wet-lab experiments. However, hundreds of resequenced human genomes (including cancerous samples) and hundreds of completely sequenced genomes of other species are accumulating at a growing pace. Thus, bioinformatic analyses of mutations are becoming feasible. It is known that mutation rates fluctuate greatly from locus to locus in mammalian genomes, and that the rates of some mutation types co-vary regionally. The causes of this variation and co-variation remain largely unexplored, and deciphering them computationally is expected to unravel the intricacies of mutagenesis. Compared with wet-lab experiments, computational analyses enable us to study mutations in their native genomic environment and on a whole-genome scale. Here we propose to utilize human resequencing data in conjunction with completely sequenced mammalian genomes in order to study regional (primarily intrachromosomal) variation and co-variation in rates of different mutation types. Additionally, we will perform comparisons between cancerous and non-cancerous mutations, highlighting the unique features of mutations associated with cancer. Several statistical multivariate and multi-scale techniques will be utilized. The proposed research will advance our understanding of mutagenesis, including the unique aspects of mutagenesis during cancer. Additionally, it will provide information vital for improving models of the evolutionary process, alignment algorithms, and algorithms for the prediction of functional elements. The tools developed here are expected to bridge interdisciplinary differences in concepts and data between biology and statistics, as well as between bioinformatics and experimental biochemistry.