PBOOST

PBOOST performs high-resolution GPU-accelerated permutation testing for genome-wide association studies (GWAS) to produce reliable P-values for SNP–SNP interaction analyses when likelihood ratio test χ² assumptions are violated due to low or zero contingency-table expected counts.

Key Features:

• GPU Acceleration: Leverages Graphics Processing Units to execute large numbers of permutations rapidly.
• Non-Parametric Approach: Uses permutation tests to avoid parametric χ² distribution assumptions of likelihood ratio tests when contingency-table expected counts are below five or zero.
• High-Speed Performance: Completes 10⁷ permutations for a single SNP pair from the Wellcome Trust Case Control Consortium (WTCCC) dataset in about one minute on an Nvidia Tesla M2090 GPU versus ~60 minutes on a CPU.
• Scalability: Tests 256 SNP pairs for 10⁷ permutations in approximately five minutes on GPU hardware compared with roughly ten hours on a CPU.
• Large-Scale Permutation Capability: Scales across a 40-node GPU cluster to perform up to 10¹² permutations across WTCCC datasets within one week, enabling validation of extremely small P-values (e.g., <1.6 × 10⁻¹²).

Scientific Applications:

• GWAS SNP–SNP Interaction Testing: Provides reliable P-value estimation for interaction analyses between single nucleotide polymorphisms in GWAS datasets such as WTCCC.
• Validation of Extremely Small P-values: Enables empirical validation of association signals with P-values beyond the resolution of parametric approximations (example threshold <1.6 × 10⁻¹²).

Methodology:

Implements non-parametric permutation testing executed on GPUs (e.g., Nvidia Tesla M2090) with the ability to run ~10⁷ permutations per SNP pair and to scale across a 40-node GPU cluster to reach ~10¹² total permutations.