Improved Imputation of Common and Uncommon Single Nucleotide Polymorphisms (SNPs) with a New Reference Set

Abstract Statistical imputation of genotype data is an important technique for analysis of genome-wide association studies (GWAS). We have built a reference dataset to improve imputation accuracy for studies of individuals of primarily European descent using genotype data from the Hap1, Omni1, and Omni2.5 human SNP arrays (Illumina). Our dataset contains 2.5-3.1 million variants for 930 European, 157 Asian, and 162 African/African-American individuals. Imputation accuracy of European data from Hap660 or OmniExpress array content, measured by the proportion of variants imputed with R^2^>0.8, improved by 34%, 23% and 12% for variants with MAF of 3%, 5% and 10%, respectively, compared to imputation using publicly available data from 1,000 Genomes and International HapMap projects. The improved accuracy with the use of the new dataset could increase the power for GWAS by as much as 8% relative to genotyping all variants. This reference dataset is available to the scientific community through the NCBI dbGaP portal. Future versions will include additional genotype data as well as non-European populations.