Many somatic mutations have been identified in cancer genomes with the ongoing cancer resequencing efforts[1].By statistical analysis,somatic mutation profile in a cancer genome can be decomposed into several mutation signatures[2],which characterize the mutation processes underlying the formation of cancer cells.It is not clear what the mutation processes are and how they are coordinated.By exploring the relationship between gene expression profile and cancer mutation profile,we aim to investigate the transcriptional regulation of somatic mutation processes.In particular,we built a latent variable model,which simultaneously group cancer patients into sub-categories and identify the latent factor that regulates mutation process.The discoveries have important implications in understanding cancer and design therapies.