Argonne researchers adapt genetics software for GPU-powered analysis of veterans’ DNA
Researchers at Argonne National Laboratory adapted a genetics analysis tool to run on graphics processing units (GPUs), after discovering that a genome-wide association study could not complete on a supercomputer’s CPUs. The work matters because it makes larger-scale analysis of veterans’ genetic data feasible, helping researchers investigate links between genetic variants and conditions such as post-traumatic stress disorder, heart disease and cancer.
The project used data from the US Department of Veterans Affairs’ Million Veteran Program, described in the article as the world’s largest and most ancestrally varied veteran biorepository. The team found that the original SAIGE software’s memory demands stalled as the cohort grew; rewriting its number-crunching component produced SAIGE-GPU, making an analysis that would have required about 481,000 CPU-core hours tractable. The article says the approach also enables researchers to search across diseases rather than starting with one condition at a time.
- GPU coding made a previously impractical genetics analysis feasible.
- The work uses data from the Million Veteran Program.
- SAIGE-GPU can support broader searches for genetic links to disease.
New here? Start with this
A genome-wide association study looks across DNA from many people to find genetic variants that occur more often alongside particular health conditions. A variant is a difference in DNA; finding an association does not by itself show that the difference causes a condition.
The Million Veteran Program, run by the US Department of Veterans Affairs, collects veterans’ health and genetic data for research. Its large and varied participant group can help researchers study conditions including post-traumatic stress disorder, heart disease and cancer.
Argonne National Laboratory researchers worked with SAIGE, software used to analyse genetic data, whose memory needs became a barrier as the dataset grew. Graphics processing units, or GPUs, can handle many calculations at once, and adapting the software may allow researchers to examine more people and a wider range of conditions.
Read the full article at the source →
Originally published by The Register as “When faster supercomputers aren’t enough”.