The Association for Computing Machinery (ACM) GBP Award Committee has named a group of researchers including MIT CSAIL postdoc Rabab Alomairy as finalists for the Gordon Bell Prize for their paper, “Toward Capturing Genetic Epistasis From Multivariate Genome-Wide Association Studies Using Mixed-Precision Kernel Ridge Regression.” The team developed software that shows how NVIDIA's GH200 Superchips can help supercomputers perform massive genetic studies more efficiently. This marks Alomairy’s second time being named a finalist for the prestigious Gordon Bell Prize, with her first recognition coming in 2022.
The award is named after and was endowed by Chester Gordon Bell, who pioneered the minicomputer and developed VAX (or Virtual Address eXtension) architecture that has helped personal computers efficiently execute complex operations. Each year since 1988, the ACM Gordon Bell Prize has recognized outstanding work in high-performance computing that uses state-of-the-art technologies.
Alomairy, a postdoc in the Julia Lab within CSAIL, specializes in high-performance computing (HPC) with a focus on its transformative potential across diverse fields like computational biology, climate science, and seismic inversion. Her research encompasses advanced numerical linear algebra, the design and optimization of scientific algorithms, the development of scalable software libraries, and task-based programming. She also applies her expertise in multicore and manycore architectures, as well as General-Purpose GPU (GPGPU) programming, to accelerate computations on modern supercomputing platforms.
Earlier this year, Rabab collaborated with colleagues from King Abdullah University of Science and Technology (KAUST) and researchers from NVIDIA and Saint Louis University. Together, the team revealed how NVIDIA's GH200 Superchips can power top-tier supercomputers to conduct large-scale genetic studies. Their software performed the largest multivariate genetic study yet, involving roughly 305,000 patients from the UK BioBank (plus over 13,000,000 patients from synthetic datasets). The team achieved FP32 accuracy with an execution rate of 1.8 EFlop/s on the Alps supercomputer from CSCS, utilizing the low precision capability of NVIDIA's GH200 Superchips.
This research is a milestone in genome-wide association studies (GWAS), which perform genotype-to-phenotype mappings, analyzing data across a genome to help identify risk factors for diseases and other traits. The team’s method shows how mixed-precision exascale computing, which processes information much faster than a standard computer, could help genome scientists efficiently analyze much larger datasets. Impressively, their software outperformed state-of-the-art software in the field by roughly 100,000 times.
Demonstrated on up to 13 million synthetic patients, it could extend GWAS to the entire population of more than half of the world’s countries and potentially help populations currently underrepresented in medical studies receive medicine more personalized to their genetic composition. It could also assist with genetically engineering grains, for example, to improve their disease resistance and show how to adapt them to a drier, hotter, and saltier world in the future.
Rabab Alomairy’s work is currently supported by an Ibn Rushd post-doctoral fellowship from KAUST. The winner of the Gordon Bell Prize will be announced at the International Conference for High Performance Computing, Networking, Storage, and Analysis in Atlanta this November.
