GeneZip: A software package for storage-efficient processing of genotype data

Abstract

Genome wide association studies directly assay 10^6 single nucleotide polymorphisms (SNPs) across a study cohort. Probabilistic estimation of additional sites by genotype imputation can increase this set of variants by 10- to 40-fold. Even with modest sample sizes (10^3-10^4), these resulting imputed datasets, containing 10^10-10^11 double-precision values, are incompatible with simultaneous lossless storage in RAM using standard methods. Existing solutions for this problem require compromises in either genotype accuracy or complexity of permissible statistical methods. Here, we present a C/C++ library that dynamically compresses probabilistic genotype data as they are loaded into memory. This method uses a customization of the DEFLATE (gzip) algorithm, and maintains constant-time access to any SNP. Average compression ratios of more than 9-fold are observed in test data.