Googling the cancer genome

Structural variants (SVs) are a particular class of mutations that have been associated to cancer formation and progression. Cancer-associated SVs can be used to determine the cancer subtype, monitor its progression and to develop novel targeted treatments. However SV analysis for personalized medicine carries many computational and algorithmic challenges. To address these challenges we developed a suite of methods for SV simulation, detection and filtering. sv-callers is a computational workflow that enable highly reproducible, portable and scalable deployment and execution of state-of-the-art SV detection algorithms across multiple high performance computing architectures. sv-gen is a workflow that can be used to generate artificial read alignment data from genomes where multiple types of SVs have been introduced at known positions. These data can serve to study how SV signals are generated at SV breakpoint positions and to benchmark SV calling methods. sv-channels is a novel deep learning-based approach for SV calling and filtering that uses one dimensional convolutional neural networks to distinguish true SVs from regions that do not contain SVs. These methods aim at improving the accuracy and cost-efficiency of SV analysis in clinical studies. This will help realize the potentials of cancer genomics for personalized medicine.