Integrating Multi-Omics into Ongoing NIH Research through Artificial Intelligence

SUMMARY
This proposal requests funds to acquire a high-performance computing (HPC) cluster with extensive data storage
and Graphics Co-Processor (GCP) capabilities for the Medical University of South Carolina (MUSC) to support
artificial intelligence (AI) based research integrating institutional laboratory, clinical, imaging and genomic data
resources. MUSC is dramatically expanding its “raw materials'' for AI research: electronic health records (EHR)
data from an expanding (now statewide) healthcare system; clinical and research-based imaging data stored in
a common format in a new PACS (picture archiving and communications system); and deep genomic data
(~100,000 over the next four years). These data create a unique opportunity along with a need for advanced
HPC capabilities to expand existing research projects using these data resources.
AI and specifically machine learning has led to groundbreaking biomedical studies, but requires HPC, robust
graphics processing units (GPUs) and large accessible storage. Currently, a number of NIH-funded investigators
at MUSC are performing research utilizing big data and machine learning, capturing massive amounts of data.
Unfortunately, these highly valuable data sets are disjointed and siloed due to a lack of an adequate, unified
data storage solution that can facilitate translational linking of these data sets.
Growing unmet computational need is also a common theme among these NIH-funded investigators, particularly
in areas of research involving high-volume, high-resolution image acquisition. Currently, inadequate access to
high performance computing and the necessary AI infrastructure limit their ability to apply techniques such as
deep learning to unlock a more complete value from these data sets. These data sets are often of a size and
complexity such that it is impractical to transmit, store and compute upon them using cloud resources, requiring
local HPC “instruments” to perform the computational tasks required. Building upon an existing MUSC
institutional initiatives, proposal adds AI computational infrastructure to the socio-technical resources available
at MUSC and to pilot level computational resources. The proposed HPC cluster will unify these efforts by
providing common data storage with ample and accessible computing power for sophisticated AI techniques
(e.g. deep learning) to exploit the potential synergies among data sets and facilitate translation efforts. The HPC
cluster, while dedicated to research, will reside in the MUSC Data Center which serves the combined MUSC
health and research enterprise.

Grant Number: 1S10OD034280-01A1
NIH Institute/Center: NIH
Principal Investigator: Alexander Alekseyenko