MPI-GWAS: a supercomputing-aided permutation approach for genomewide association studies.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Permutation testing is crucial for accurate significance testing in genomic research, particularly in genome-wide association studies (GWAS).
• High computational cost of permutation testing poses a significant challenge for large-scale GWAS.
• Existing methods struggle with the intensive computational demands of genome-wide permutation analyses.

Purpose of the Study:

• To develop and evaluate a supercomputing-aided approach for accelerating permutation testing in GWAS.
• To reduce the computational burden of permutation testing while maintaining accuracy in identifying significant genetic associations.
• To enable feasible and timely execution of permutation-based GWAS on large datasets.

Main Methods:

• Developed MPI-GWAS, an application utilizing the Message Passing Interface (MPI) on a parallel computing architecture.
• Leveraged the Nurion supercomputer (8,305 nodes, 563,740 CPUs) for high-performance computing.
• Performed large-scale permutation testing for GWAS, analyzing millions of permutations across thousands of subjects and loci.

Main Results:

• MPI-GWAS significantly accelerated permutation testing, completing 10^7 permutations for ~30,000-50,000 loci in over 7,000 subjects in approximately 4 days.
• Demonstrated efficient parallelization of permutation testing using 2,720 CPU cores for 10^7 permutations of a single locus in 600 seconds.
• The supercomputing approach effectively harnesses parallel computing resources to make permutation-based GWAS computationally feasible.

Conclusions:

• MPI-GWAS provides a computationally efficient solution for permutation testing in GWAS.
• The developed approach enables timely and accurate significance testing in large-scale genomic studies.
• Harnessing supercomputing power with MPI-GWAS overcomes previous computational limitations in GWAS.