A High-Performance Computational Platform for the Brain and Data Sciences

University Of Virginia, Charlottesville VA

Investigators

Abstract

Neuroscience research “at scale” is both enabled and challenged by advances in technology and data science. Advancements in the acquisition of neuroimaging, electrophysiological, genetic, and other -omic biomedical data types have also created novel interdisciplinary studies in the field of neuroscience and biomedical informatics. Rapid innovations in neuroscience computation and data science have resulted in the creation of massive amounts of quantitative results concerning the human brain. Researchers at the University of Virginia (UVA) and across the U.S. increasingly rely on such computationally intensive methods to conduct data-driven studies at the levels of cells-to-systems-to-syndromes-to-societies. Doing so necessitates access to advanced, specialized cyberinstruments specially designed for high-throughput neuroscience research workflows. However, investigator access to high-end computing instruments at UVA has been inadequate and piece-meal, resulting in defaulting to small-scale, homegrown laboratory-based solutions. Such constraints have resulted in both a resource bottleneck as well as a non-optimal workflow for neuroscience research at UVA. To support neuroscience discovery at scale, a specialized, well-integrated high-performance computing instrument is needed to advance data analytics and machine learning applied to the available and growing collection of brain data types. In response to these computational challenges, the UVA-based community of neuro-, biomedical, and data scientists with common interests in brain data and its computational needs seek to acquire a shared, neuroscience-dedicated, and much-needed instrument. The new cyberinstrument, called “NEO” - powered by 32 combined NVIDIA “Grace Hopper” GPU “Superchip” nodes - will exist at the very heart of UVA’s NeurosciencE discOvery at Scale (NEO@scale) Program. The requested, leading-edge, GPU-driven system will profoundly enhance UVA’s brain data analysis and modeling capabilities, expand data processing workflow bandwidth, permit the application of cutting-edge machine/deep-learning and AI methods, and form an essential tool in data visualization. Specifically designed to out-price and out-perform cloud-based services, this instrument will permit in a many-fold speed up over currently available resources. With impressive institutional support, NEO will support the laboratories of 28 NIH-funded UVA researchers. A formal administrative plan, building upon UVA’s established research computing collaborative, will assure that the equipment is effectively managed. UVA technical and management personnel will provide 24/7 availability as well as tailored user training activities. Application areas for targeted advanced data analytics using NEO include large-scale human neuroimaging, epigenetics, epilepsy, focused ultrasound, molecular/cellular processes, etc. The deployment of the proposed computational instrument is directly in line with UVA’s stated institutional goals for advancing research excellence in brain and data sciences - meeting the large-scale computational and quantitative demands needed for accelerating understanding of brain form and function in health and disease.