The Bioconductor project is rooted in recognition that ef?cient, rigorous, and reproducible analysis of high- dimensional data can be achieved when statisticians, biologists, and computer scientists federate efforts in a transparent and carefully engineered way. The project Accelerating Cancer Genomics with Cloud-scale Bio- conductor devises new approaches to carrying out genome-scale analysis of cancer data using cloud computing environments. The proposal is based on strategies that have proven highly effective in ?fteen years of supporting collaborative and carefully engineered software for genome scale analysis in computational biology in the Biocon- ductor project, based on the highly portable and widely adopted R language and environment for data analysis.
In Aim 1 we develop architecture and infrastructure for scalably harvesting cloud-based representations of large- scale cancer genome studies such as The Cancer Genome Atlas, creating formal high-performance work?ows for processing and interpreting cancer genome analyses, and providing packaging and data distribution schemes for moving data to the cloud for scalable analysis there.
In Aim 2 we create and support independent creation of intuitive and cancer-relevant interface components supporting reproducible interactive exploration and analysis using the facilities of Rstudio.
In Aim 3 we update and generalize the Bioconductor MLInterfaces metapackage to support advanced machine learning using the cancer-oriented strategies and facilities devised in Aims 1 and 2. Our proposal will bene?t large numbers of cancer researchers who will be taking advantage of cloud resources, probably with R close to hand, by marrying strengths of cloud-centric strategies for data archiving and query resolution, to the strengths of Bioconductor development and analysis capabilities. We have letters of support from the leadership of the three NCI Cancer Cloud Pilot projects for this project.
Despite major advances in elucidating mechanisms of tumor initiation and proliferation, treatment strategies for many cancers are ineffective, and patient-to-patient variation in treatment response suggests that personal targeting of cancer based on tumor molecular pro?les will be necessary. This proposal takes a design and architecture approach from a widely used project for analyzing general data arising in genome-scale biology, and adapts it to new NCI-supported cloud-based data archives and analysis environments. The proposal will accelerate identi?cation of sources of variation of tumor responsiveness to treatment and will aid physicians in devising personalized antitumor strategies.
