LAUSR

Abstract When diagnosed at an advanced stage, most cancer patients suffer from treatment failure, recurrences and low survival. Taking advantage of high-throughput sequencing and deep learning techniques, we developed an early cancer monitoring method based on multi-modal deep Boltzmann machine to (1) learn association between matched germline and somatic mutations captured by whole exome sequencing from available samples of cancer patients, and (2) predict patient-specific high-risk genes whose somatic mutations are required to drive normal tissues to a tumor state. Our experiments on a set of breast cancer samples show that our method significantly outperforms the currently used frequency-based method in the personalized prediction of genes carrying critical mutations.