permGWAS

permGWAS implements permutation-based significance thresholds for genome-wide association studies to improve statistical rigor in detecting genotype-phenotype associations.

Key Features:

• Python implementation: Implemented in Python for genome-wide association analysis.
• LMM reformulation with 4D tensors: Reformulates linear mixed models (LMM) using 4D tensors to enable efficient computation of permutation-based significance thresholds.
• Multi-CPU and GPU acceleration: Supports multi-CPU and GPU execution to reduce computational time for large-scale permutation testing.
• Demonstrated scalability: Demonstrated re-analysis of over 500 Arabidopsis thaliana phenotypes with 100 permutations each in under eight days on a single GPU.
• Improved statistical control: Uses permutation-based thresholds to provide more realistic significance levels and lower false discovery rates compared to Bonferroni correction, particularly for skewed phenotypic distributions.

Scientific Applications:

• Plant genetics (Arabidopsis thaliana): Applicable to large-scale GWAS in plant genetics, including analyses of Arabidopsis thaliana phenotypes.
• Permutation-based GWAS significance estimation: Provides empirical significance thresholds for GWAS where LMM assumptions such as normal residuals and marker independence may be violated.
• Large-scale phenotype analyses: Suited for extensive multi-phenotype studies requiring many permutations to control false discoveries.

Methodology:

permGWAS reformulates linear mixed models using 4D tensors to perform permutation testing and to compute permutation-based significance thresholds, with support for multi-CPU and GPU execution.