IHT

IHT performs sparse penalized regression via iterative hard thresholding to jointly analyze genome-wide SNP data across multiple correlated traits for multivariate GWAS.

Key Features:

• Penalized regression algorithm: Models all single nucleotide polymorphisms (SNPs) jointly across multiple traits within a single multivariate regression using a sparse penalized regression framework.
• Iterative hard thresholding methodology: Applies iterative hard thresholding to iteratively retain significant predictors and discard others, reducing model dimensionality and computational load.
• Performance efficiency: In simulation studies up to 100 traits, demonstrated competitive true positive rates and reduced false positive rates compared with GEMMA's linear mixed models and mv-PLINK's canonical correlation analysis, with faster execution times.
• Scalability and implementation: Implemented in the Julia package MendelIHT.jl and reported to scale to large datasets (e.g., ~20 hours for a 3‑trait joint analysis and ~53 hours for an 18‑trait joint analysis on UK Biobank data using up to 80 GB RAM).

Scientific Applications:

• Multivariate GWAS: Jointly analyze multiple correlated traits to identify SNPs with shared or trait-specific effects.
• Genetic architecture analysis: Investigate complex trait interactions and polygenic influences by modeling shared genetic effects across traits.
• Large-scale genetic analyses: Apply sparse multivariate association mapping on datasets with many traits and genome-wide SNP coverage.

Methodology:

IHT applies a sparse penalized regression framework using iterative hard thresholding to select a subset of SNP predictors within a joint multivariate regression across traits.