DOE and NIH Partnerships -- Cancer and Brain
Argonne National Laboratory; University of Chicago
Department of Energy laboratories have an unprecedented capability to apply computational methods to problems of national importance. Two national initiatives recently have emerged that are ideally suited to leverage DOE computational science via a partnership with the National Institutes of Health. The first is the U.S. BRAIN Initiative, with goals to classify neural cell types, map brain circuits, determine functional brain maps, advance neuroscience and translate research results into biomedical and clinical practice. I will discuss the connectome project, aimed at developing methods to reverse-engineer the "wiring" diagram of brains. Creating high-resolution static pictures of the brain requires high-throughput imaging, large-scale data analysis, novel database structures, machine-learning methods for segmentation and classification, and, ultimately, large-scale computing for modeling the resulting networks. To further the goal of producing connectomes for a diverse set of organisms, a National Brain Observatory would be established as a user facility for on-demand neural circuitry reconstruction problems.
The second national effort is the U.S. Cancer Moonshot initiative to advance cancer research and personalized medicine. DOE is collaborating with the National Cancer Institute to pilot three problem areas involving large-scale computing. The first is large-scale modeling and simulation of the RAS/RAF proteins involved in 30 percent of all cancers. The project aims to discover new drug targets in molecular pathways associated with the RAS oncogene. Second is the development of predictive cancer drug response models based on the molecular properties of patient tumors. This will enable development of personalized treatment strategies that match drugs with patients and tumors. The third area is mining the national cancer surveillance database with large-scale data analysis techniques to design better treatment strategies for the U.S. population. All three of these problems require large-scale machine learning and target exascale computing.