Iterative hard thresholding for model selection in genome-wide association studies

A genome-wide association study (GWAS) correlates marker and trait variation in a study sample. Each subject is genotyped at a multitude of SNPs (single nucleotide polymorphisms) spanning the genome. Here, we assume that subjects are randomly collected unrelateds and that trait values are normally distributed or can be transformed to normality. Over the past decade, geneticists have been remarkably successful in applying GWAS analysis to hundreds of traits. The massive amount of data produced in these studies present unique computational challenges. Penalized regression with the ℓ₁ penalty (LASSO) or minimax concave penalty (MCP) penalties is capable of selecting a handful of associated SNPs from millions of potential SNPs. Unfortunately, model selection can be corrupted by false positives and false negatives, obscuring the genetic underpinning of a trait. Here, we compare LASSO and MCP penalized regression to iterative hard thresholding (IHT). On GWAS regression data, IHT is better at model selection and comparable in speed to both methods of penalized regression. This conclusion holds for both simulated and real GWAS data. IHT fosters parallelization and scales well in problems with large numbers of causal markers. Our parallel implementation of IHT accommodates SNP genotype compression and exploits multiple CPU cores and graphics processing units (GPUs). This allows statistical geneticists to leverage commodity desktop computers in GWAS analysis and to avoid supercomputing.

Source code is freely available at https://github.com/klkeys/IHT.jl.